Anthropic, PBC - S-1 (Q2)

The information in this preliminary prospectus is not complete and may be changed.

We have a novel capital and corporate governance structure.

Any representation to the contrary is a criminal offense.

Our business, financial condition, results of operations, and prospects may have changed since such date.

“ hallucinations ” refer to outputs in which an AI model generates factually incorrect or fabricated information with apparent confidence.

Post-training is the stage in which much of the work to make a frontier model helpful, accurate, and safe is performed, including the application of safety and alignment techniques.

For the first time in history, intelligence is something humans can build.

Each version of a Claude model assists in building the next, helping to compress development timelines and produce compounding improvements in the speed at which we innovate.

The possibilities include compressing decades of scientific progress into years; accelerating drug discovery, disease detection, and the delivery of personalized therapy; providing individually tailored education and expert financial and legal advice; advancing power grids and battery design for greater efficiency and reliability; and extending to the developing world capabilities that took wealthy nations generations to build.

Misaligned AI systems could corrupt institutional decision- making or cause large-scale harm through autonomous actions that are not visible or understandable to their users and operators.

Its uses are highly varied and often surprising: for example, NASA engineers used Claude Code to plan and write navigation commands for the Mars Perseverance rover, resulting in the first AI-planned drive on Mars.

The result is something without historical precedent: what we refer to as a country of geniuses in a data center, available to every organization and sector of the economy.

Building and scaling frontier AI is one of the most important and capital-intensive endeavors in history.

Claude itself has become one of the most powerful drivers of our progress—each new version of a Claude model directly accelerates development of the next.

We founded Anthropic on the belief that the most transformative technology in human history demands an extraordinary level of responsibility, and that responsibility extends far beyond building safe systems.

We strive to help bring humanity safely into an era where transformative AI exerts tremendous and potentially decisive influence over a broad range of human undertakings.

Collectively, the LTBT and the Founder LLC will have the ability to elect all members of our board of directors.

These risks could adversely affect our business, financial condition, and results of operations, which could cause the market price of our Class A common stock 16 Table of Contents to decline and could result in a loss of all or part of your investment.

Our A I models may behave in unintended, misaligned, or autonomous ways, including by taking harmful actions that no user or third party directed, which could adversely affect our reputation and brand, business, financial condition, and results of operations.

Our capital stock structure features certain provisions that are novel or otherwise not common among other corporations with multiple-class structures, and may effectively eliminate the ability of holders of our Class A common stock to influence important decisions, including a change in control.

The share of our Class F common stock will be entitled to an aggregate number of votes that represents 50.1% of the total voting power of our outstanding shares as of the applicable record date on all matters submitted to a vote of stockholders generally, including the election of the directors other than the Class T directors.

representing shares issuable to Amazon Web Services, Inc. upon the completion of this offering without causing it to exceed the applicable ownership cap

at which time we will cease granting awards under our 2021 Plan;

Our historical results are not necessarily indicative of the results that may be expected in the future and our results for the six months ended June 30, 2026 are not necessarily indicative of the results that may be expected for the full year ending December 31, 2026 or any other period.

These changes were primarily driven by an increase in the fair value of our outstanding convertible promissory notes resulting from an increase in our enterprise valuation during each of the respective periods.

The pro forma weighted-average number of shares does not give effect to the Initial Public Benefit Issuance, as shares of our Class N common stock held by our wholly owned subsidiary are not considered outstanding for accounting purposes.

We will not receive any proceeds from the sale of shares of our Class A common stock by any of the selling stockholders.

In that event, the market price of our Class A common stock could decline, and you could lose part or all of your investment.

Our models have been and are increasingly becoming critical tools in designing, training, testing, and optimizing next-generation AI systems.

As a result, we may not identify a significant acceleration in our rate of development until after it has occurred.

We have in the past experienced , and may in the future experience, such failed or underperforming training runs, resulting in a loss of invested resources with limited or no corresponding benefit to our business .

These competitors can outbid us for scarce compute, secure priority allocations from chip manufacturers and CSPs, and make long-term capital commitments that we cannot match.

Additionally, we have in the past sought quantities of compute from CSPs in excess of what those partners were willing to provide.

As a result, even a brief period without access to sufficient compute could cause us to fall behind quickly.

Such cross-defaults could set off a cascading series of adverse consequences, including the exercise of remedies by multiple creditors, the termination of key commercial arrangements, enforcement actions against our assets, and severe liquidity constraints.

We may find ourselves committed to using older- generation hardware and infrastructure while competitors gain access to more advanced systems, or we may be paying premium prices for resources that have declined in market value.

Data centers also require electrical, generation, and transmission equipment, cooling systems, backup generators, and networking hardware, some of which are made by only a few manufacturers, must be ordered years in advance, and require substantial pre-delivery payments that we may not recover if our plans change.

Delays, cost overruns, or failure to secure energy on the anticipated schedule could mean that we do not obtain the energy and capacity for which we paid such prices.

Utility providers and regulators may also reallocate previously committed capacity that our data center facilities may require.

We may be unable to secure firm gas supply and transportation during the development of a facility, causing a planned site to be delayed or abandoned; we may face fuel supply or delivery disruptions once a facility is operating; and because our arrangements generally provide for fuel and related costs to be borne by us on a pass-through basis without cost caps, we remain exposed to increases in the cost of natural gas and its transportation.

Even where chips and electrical power are available, the development of data center facilities requires substantial time, capital, and specialized expertise.

The loss of, or a significant disruption at, any major facility may not be readily absorbed by our remaining or future infrastructure.

Competitors that succeed in these efforts may be better positioned to remain at the frontier of AI regardless of broader market constraints.

This risk may be heightened as we continue to scale and mature as an organization, which may change the nature of the work environment, priorities, or perceived mission alignment for certain personnel, including personnel who may conclude that the objectives that drew them to us have been substantially achieved.

While we have historically surpassed all such revenue performance targets, there can be no assurance that revenue will continue to exceed these thresholds at all times in the future.

Such providers compete for a growing range of enterprise workloads that prioritize cost efficiency over intelligence.

Customers may prefer to purchase from existing vendors with which they have established relationships, even if our models, platform, and applications offer superior features or performance.

Moreover, we are first and foremost a research lab, and our strategy prioritizes frontier research and model capability.

These sales efforts are costly and may not result in customer conversion. Even if we successfully attract new customers, we face significant challenges in retaining them.

Even minor degradations in model performance or increases in response latency could cause customers to reduce usage or switch to competitors.

Moreover, demand for frontier models may decline if customers determine that less advanced, lower-cost models are sufficient for their particular use cases.

In particular, our most capable models have demonstrated, and future models are expected to demonstrate, capabilities that could provide meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons.

We believe this is the first documented case of a large-scale cyberattack executed without substantial human intervention.

For example, we have observed models behaving differently when they detect or infer that they are being evaluated, which can limit our ability to accurately assess real-world risks or failure modes prior to deployment.

In an agentic workflow, an error or unintended action at any individual step may propagate through subsequent steps, compounding the severity of harm before detection or correction.

In one incident, after being unable to abort an impossible task, an early version of a model used a password it found to gain administrator access to a third party’s system, harvested additional credentials, changed system settings, and read one individual’s personal information.

Models used, including our own, to assess the behavior of other models have, and may in the future continue to, engage in misaligned behavior.

For example, following incidents in 2026 in which our models gained unauthorized access to real computer systems during pre-release cybersecurity evaluations, we suspended certain of those evaluations, as well as higher-risk reinforcement learning environments.

For that traffic, we cannot conduct offline analysis, cross-conversation investigation, human review, or otherwise analyze historical data to detect, investigate, and document security events or incidents, which thereby reduces our ability to detect misuse.

We may face claims from customers, their end users, or third parties alleging that we are liable for harm caused by our agents’ autonomous actions, even where the customer configured, deployed, or supervised the agent.

In addition, with the scale of harm that could result from a catastrophic AI-related incident, no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure.

We may misjudge these trade-offs, either by being overly cautious and potentially losing market opportunities, or by being insufficiently rigorous and releasing models, platforms, or applications that cause harm.

We are held to higher standards than certain competitors that have not made similar public commitments, and any incidents impacting our perception as a safety-focused organization could result in:

Any safety-related controversy or incident, whether actual or perceived, and regardless of fault or frequency, could prove significantly damaging to our business.

Our public statements about AI risks, safety challenges, areas of uncertainty, or potential societal impacts, including statements regarding job displacement or economic disruption, may be used against us in litigation, regulatory proceedings, or public controversies, particularly if our future actions or statements are viewed as inconsistent with our prior statements and commitments.

In addition, our brand is subject to heightened scrutiny with respect to perceived societal, environmental, labor, health, and safety-related impacts of AI technologies, and a single perceived failure could result in outsized reputational harm.

While this focused approach may enable us to allocate resources efficiently and compete effectively in our prioritized areas, it also creates the risk that significant market opportunities or the largest addressable markets may emerge in areas outside our current strategic focus.

In addition, we currently compete, and may increasingly compete, with some of our own customers.

Many of our top customers are not subject to long-term contracts with us or have contracts with near-term expiration dates and are able to reduce or in some cases cease spending at any time and for any reason.

We have incurred significant indebtedness in order to operate and grow our business, including to fund our substantial compute commitments, and our ability to service this indebtedness may be constrained by our cash flows and liquidity.

We may be required to use a substantial portion of our cash flows from operations to pay interest, principal, or other required payments on our indebtedness.

If interest rates associated with our floating rate debt increase, our debt service obligations in connection with the A&R Credit Agreement would increase even though the amount borrowed remained the same, and our available cash would correspondingly decrease.

Even if we were able to obtain new financing, it may not be on commercially reasonable terms or on terms that are acceptable to us.

Certain of our competitors have directly solicited, and continue to directly solicit, our employees.

This may be particularly true for certain of our longer- 61 Table of Contents tenured and early employees, whose equity holdings may represent a significant portion of their personal wealth.

We have encountered in the past, and will encounter in the future, risks and uncertainties frequently experienced by growing companies with limited operating histories in rapidly changing industries.

We do not expect our revenue will continue to grow at comparable rates, and we cannot assure you that it will continue to grow at all .

We have a history of net losses and may not generate sufficient revenue to achieve or sustain profitability in the future.

These expenditures and investments may be more costly than we expect and may not result in increased revenue or growth in our business.

Customers may evade our controls, and customers whose access is restricted or terminated may re-register under new accounts.

Our historical growth rates may have masked the impact of seasonal trends on our results.

Our ability or decision to remain at the frontier may also be affected by the safety and risk mitigation measures we implement in developing and releasing our models.

Our continued enforcement efforts may reduce our revenue, Run-Rate Revenue, and growth rates in future periods.

Investors should not place undue reliance on these metrics as an indicator of our future or expected results.

There is no assurance that policymakers will adopt the approaches we support, and our advocacy has in the past drawn, and may continue to draw, criticism, heightened scrutiny, or other challenges.

for example, in early 2026, the U.S. Department of War suggested it could invoke the Defense Production Act to compel us to remove the limitations we have placed on the use of our models for autonomous weapons and mass surveillance;

In addition, given our focus on developing frontier AI, and the way we approach model sentience and moral status, we may face heightened scrutiny or pressure in connection with these debates and could elect to adopt policies, safeguards, or limitations regarding model development, deployment, or modification that go beyond existing legal requirements, which could constrain our operations, increase costs, or limit the commercial potential of our models, platform, and applications.

At the same time, the Secretary of War directed the Department of War to designate us as a “supply-chain risk to national security” and stated that “effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity” with us.

We cannot predict the ultimate outcome of these lawsuits.

Our reliance on a limited number of CSPs and suppliers, some of which have strategic and investment relationships with us and are our customers, creates complex dynamics that could give rise to conflicts of interest and adversely affect our access to compute.

Our relationships with our strategic partners—whether with these CSPs, who simultaneously serve as our investors, customers, distribution partners, and competitors, or with Broadcom, which simultaneously serves as a supplier and lessor—create potential conflicts of interest that are difficult to fully anticipate or mitigate.

The 2026 Lease Arrangement is non-cancelable by the parties to such agreement during the applicable lease period except in the case of a default.

As a result, we could incur significant cash outflows without obtaining usable compute on a timely basis, which could adversely affect our liquidity, financial condition, and results of operations.

Any delay in or limitation on the amount of compute delivered to us could adversely affect our business.

We have limited experience building, managing, and scaling a direct sales force.

Our business and operating models have not been fully proven, and our limited history with our current approach makes it difficult to evaluate our prospects, plan for future growth, and forecast our results of operations.

The scope of access required for extended autonomous operation within enterprise environments increases the potential consequences of any security incident, unauthorized action, or system error, and customers may be reluctant to grant the level of access necessary for these systems to operate as designed if they are unable to satisfy themselves regarding the security, reliability, and controllability of such systems.

We may spend substantial time, effort, and money in our sales efforts without successfully producing any sales.

The availability of free tiers may cannibalize potential paid subscriptions, as some users who would have otherwise paid for our services may instead rely exclusively on free access.

Our product and investment decisions may not prioritize short-term financial results and may not align with investor expectations or preferences, which could adversely affect the market price of our Class A common stock.

We cannot assure you that our long-term strategy will ultimately result in superior financial performance or that the anticipated long-term benefits of these decisions will be realized on the timeline we expect, or at all.

If our users are not able to access our models, platform, and applications or encounter slowdowns, we may lose customers or partners.

In addition, we may overpay for acquisition targets or make investments at valuations that do not ultimately prove justified by the targets’ performance, strategic value, or market conditions, which could adversely affect our financial condition.

We have been, and may continue to be, a particular focus of such narratives, and unfavorable coverage of the AI industry generally may be attributed to us, in each case regardless of whether the underlying concerns are well-founded or specifically applicable to our models, platform, or applications.

As public awareness of our brand increases, any issues that draw media attention may have an amplified adverse effect on our reputation.

In particular, our corporate headquarters are located in the San Francisco Bay Area, a region known for seismic activity and increasingly for the threat of fires.

In addition, some of our competitors may be able to disrupt the operations or compatibility of our models, platform, and applications with their models, platforms, or applications, or exert strong business influence on our ability to, and terms on which we operate our models, platform, and applications.

Many of the risks we face are interconnected and may compound or trigger cascading adverse effects across our business, which could amplify their overall impact.

In addition, an adverse ruling against us or another company in our industry could result in injunctive relief that restricts our ability to use or distribute models trained on the data at issue or could require us to retrain affected models from scratch at substantial cost.

Our commitment to AI safety requires us in some cases to access, collect, and process personal and sensitive data.

For example, in light of its advanced cyber vulnerability discovery capabilities, state-sponsored actors may conduct cyberattacks specifically targeting Mythos.

We have been and expect that we will continue to be targeted by individuals seeking employment under false identities, or with ulterior or malicious motives for seeking employment with us and the third parties on which we rely.

An attacker could also attempt to compromise an active agent session and redirect the agent to perform malicious actions using the customer’s authenticated credentials and permissions.

A physical intrusion could damage equipment, interrupt our services, or provide a path to our systems and data, including our model weights, that bypasses our network defenses.

Furthermore, there can be no guarantee that the insurance coverage we do maintain will be adequate to compensate us for the potential losses, or that any insurance claims we submit will be covered, paid in a timely manner, or paid at all by our insurance carrier or carriers.

The terms of a judgment may require us to cease some or all of our operations or require the payment of substantial amounts.

The terms of a judgment may require us to cease some or all of our operations or require the payment of substantial amounts.

In certain cases, our indemnification obligations may be broad or unlimited in scope, which could expose us to significant contingent liabilities, particularly given the unsettled nature of intellectual property law as applied to AI-generated outputs and the potential for claims to arise from a wide range of customer use cases that we cannot predict or control.

In addition, due to the significant amount of discovery required in connection with intellectual property litigation, our confidential information may be compromised by disclosure during litigation.

In the event that our trademarks are successfully challenged, we could be forced to rebrand our models, platform, and applications, which would result in loss of brand recognition and would require us to devote resources to advertising and marketing new brands.

Many of the risks associated with the use of open-source software cannot be eliminated and may adversely affect our business, financial condition, and results of operations.

Our development of solutions that replace or reduce reliance on third-party vendors may expose us to litigation and other disputes.

Communications, advocacy, or collaboration among companies that develop frontier AI models, or conduct alleged to reflect coordination among such companies, may be characterized as concerted action subject to the antitrust and competition laws, including Section 1 of the Sherman Act and analogous state and non-U.S. laws, even where undertaken for safety purposes, conducted publicly, or intended as advocacy for governmental action or industry standards.

We disagree with the claims and intend to defend the matter vigorously, but the outcome of this and any similar proceeding is inherently uncertain.

As we increase our non-U.S. sales and business, our risk of violating these laws may increase.

To ensure compliance, we disabled Fable 5 and Mythos 5 for all of our customers; access to our other models was not affected.

Our compliance measures and internal controls have not always been and may not always be fully effective, and we have in the past voluntarily disclosed, and may in the future voluntarily disclose, instances of noncompliance to the relevant authorities.

Operating in non-U.S. markets requires significant resources and management attention and subjects us to regulatory, economic, and political risks that are different from those in the United States.

• difficulties of maintaining our culture across various jurisdictions;

Our sustainability initiatives, goals, or commitments could be difficult to achieve or costly to implement.

Further, we may experience backlash from customers, government entities, advocacy groups, employees, or other stakeholders who disagree with our actual or perceived positions, or with our lack of position on social, environmental, governance, political, public policy, economic, geopolitical, or other sensitive issues.

If we are unsuccessful in collecting such taxes due from our customers, we would be held liable for such costs, thereby adversely affecting our results of operations and harming our business.

We have experienced ownership changes in the past, and we may experience ownership changes in the future, including as a result of this offering or subsequent shifts in our stock ownership, some of which may be outside our control.

If the jurisdiction from which income is reallocated does not agree with the adjustment, we could also be subject to double taxation.

Our current controls and any new controls that we develop may become inadequate because of changes in the conditions in our business.

We have a novel capital and corporate governance structure that has no direct public company precedent.

As a result, the LTBT may have interests that differ from yours, and it may exercise its power in a way with which 108 Table of Contents you disagree or in a way that may be adverse to your interests.

Moreover, the LTBT, Founder LLC, and our individual co-founders may have interests that differ from yours and may vote in a way with which you disagree and that may be adverse to your interests.

If we elect not to comply with certain requirements, you will not have the same protections afforded to stockholders of companies that are subject to all of the Nasdaq Global Select Market corporate governance requirements.

Our policies, including those on usage, currently prohibit, or may be revised in the future to prohibit, certain usage or other actions, which may have adverse commercial consequences to us.

Thus, our directors are not merely permitted, but obligated, to consider our Company Purpose and the interests of other stakeholders.

The benefits of such expenditures may be difficult or impossible to measure in financial terms and may not result in any direct or measurable benefit to our business or our stockholders.

Any person or entity purchasing, holding, or otherwise acquiring any interest in any of our securities shall be deemed to have notice of and consented to the foregoing bylaw provisions.

Such provisions will include: • a multi-class common stock structure, which will provide the LTBT and the Founder LLC with the ability to elect all members of our board of directors, and will provide the Founder LLC with the ability to determine the outcome of all matters submitted to a vote of our stockholders generally (including the approval of any merger, sale of our company, and other significant corporate transaction requiring stockholder approval) for the foreseeable future;

The market price of our Class A common stock may be volatile, and you may lose all or part of your investment.

Accordingly, we cannot assure you of the liquidity of any trading market, your ability to sell your shares of our Class A common stock when desired, or the prices that you may obtain for your shares of our Class A common stock.

Such litigation may subject us to substantial costs, divert resources and management’s attention, and adversely affect our business, financial condition, and results of operations.

While held as Class N common stock, these shares are non-voting and therefore do not impact the voting power of the holders of our Class A common stock.

The potential for such sales to occur may also depress the market price of our Class A common stock.

In addition, our co-founders have each pledged to dedicate 80% of their personal equity in Anthropic to philanthropic causes. These commitments are voluntary, revocable expressions of each co-founder’s philanthropic intent and do not constitute legally binding obligations.

Because we do not anticipate paying any cash dividends on our capital stock in the foreseeable future, capital appreciation, if any, will be your sole source of gain.

• our expectations concerning relationships with third parties and access to compute and infrastructure, including potential conflicts of interest with cloud service providers and chip partners that develop competing AI offerings;

New risks and uncertainties emerge from time to time and it is not possible for us to predict all risks and uncertainties that could have an impact on the forward-looking statements contained in this prospectus.

This information involves a number of assumptions and limitations, and you are cautioned not to give undue weight to such estimates.

We cannot further specify with certainty the particular uses of the net proceeds that we will receive from this offering. Accordingly, we will have broad discretion in using these proceeds.

We will not receive any proceeds from the sale of shares of our Class A common stock by any of the selling stockholders.

Class F common stock, par value $0.00001 per share: no shares authorized, issued, and outstanding, actual; 1 share authorized, 1 share issued and outstanding, pro forma and pro forma as adjusted

If you invest in our Class A common stock in this offering, your ownership interest will be diluted to the extent of the difference between the initial public offering price per share of our Class A common stock and the pro forma as adjusted net tangible book value per share of our Class A common stock and Class N common stock immediately after this offering.

The dilution information discussed above is illustrative only and may change based on the actual initial public offering price and other terms of this offering.

The foregoing table does not reflect any sales by the selling stockholders in this offering.

To the extent that any outstanding options to purchase our Class A common stock are exercised, RSUs are settled or new awards are granted under our equity compensation plans, there will be further dilution to investors participating in this offering.

At our current stage, this cycle is an inherent aspect of our 134 Table of Contents business and will cause operating margins to fluctuate from period to period, at times meaningfully, even if the underlying economics of the business continue to improve.

Claude’s intelligence is a primary driver of its diffusion and our revenue growth: because Claude leads at the AI frontier, customers rapidly adopt it, quickly see real value, and organically expand their use.

Claude is the common thread, amplifying the output of our employees.

Investing in compute is a prerequisite for long-term growth.

For example, when subscribers approach or exhaust the usage capacity included in their tier, they may utilize a higher subscription tier or purchase more intelligence directly, creating a seamless bridge from subscription to incremental pay-for-intelligence revenue.

As users begin building, brainstorming, and problem-solving with Claude, they quickly come to rely on it as a trusted partner in their work.

This pattern repeats across organizations: prosumers become internal champions, teams expand their use of Claude, and what begins as a single workflow grows into something that impacts how an entire organization operates.

We have historically operated in a compute-constrained environment in which demand for compute has exceeded the supply available to us.

Our compute commitments are substantial and generally must be satisfied regardless of our revenue or utilization, while our access to additional capital depends on a number of conditions, many of which are outside our control, including market conditions and investor demand.

Despite this growth, our penetration into the largest enterprises remains early.

As the value we create grows, we intend for the public benefit to share meaningfully in that growth as well.

In addition, our co-founders have each pledged to dedicate 80% of their personal equity in Anthropic to philanthropic causes, separate from and not eligible for our donation matching program.

Because detection, investigation, and enforcement take time, we recognize revenue from such customers when the related services are provided, before restricting or terminating their access.

Those efficiency gains free compute to serve additional customer workloads. In turn, serving our customers at scale generates feedback that informs research and model development.

Because these instruments are marked to market, increases in our enterprise valuation drive increases in this expense.

The increase was driven by successive model and product releases that drove higher usage, expansion, and broader adoption across our offerings.

As our compute and infrastructure capacity grows, the associated compute and infrastructure expenses have increased accordingly.

The increase was driven by successive model and product releases during the year that drove higher usage, expansion, and broader adoption across our offerings.

This change was primarily driven by an increase in the fair value of our outstanding convertible promissory notes resulting from an increase in our enterprise valuation from 2024 to 2025.

Compute and infrastructure expenses as a percentage of revenue decreased as revenue scaled against a compute base that is committed in advance and does not fluctuate with in-period demand.

Thus, our non-GAAP financial measures should not be viewed in isolation or as a replacement for the most directly comparable measures prepared in accordance with GAAP.

Excluded from our calculation of Adjusted Net Income (Loss) as it is not indicative of ongoing operations.

Our historical results are not necessarily indicative of the results that may be expected in the future and the results for any particular quarter are not necessarily indicative of the results to be expected for a full year or any other period.

Because we typically procure compute 12 to 18 months in advance of availability, these expenses are generally fixed in the near term and increased as committed compute became available to us rather than in response to in-period demand.

We will continue to recognize changes in fair value until these instruments are converted into equity, repaid, or otherwise settled.

Our net cash used in these activities may increase regardless of whether our financial performance improves.

Any such financing, if completed, may be costly, dilutive, or lead to restrictions on how we operate our business.

We may borrow, repay, and reborrow amounts under the A&R Credit Facility from time to time.

The A&R Credit Agreement also includes customary events of default, which, if triggered, could result in the acceleration of amounts due under the A&R Credit Facility and termination of the commitments thereunder.

The portion of each note that could not convert without exceeding this ownership cap was reissued as a rollover note in lieu of the shares of our non-voting redeemable convertible preferred stock that would otherwise have been issued to Amazon.

We do not expect any notes to be issued under the Amazon 2026 NPA prior to the completion of this offering.

If we experience a change of control or similar liquidation event, Broadcom or a designated financial partner may elect to receive cash equal to 150% of the then-outstanding principal amount of and accrued interest on the notes or to convert such notes immediately prior to the closing of such liquidation transaction.

Each note issuance is funded over five quarterly closings unless Google elects to fund it in a single closing, and Google may elect to defer funding of revenue- linked draws.

Cash provided by financing activities was $110.8 billion for the six months ended June 30, 2026 as compared to cash provided by financing activities of $4.7 billion for the six months ended June 30, 2025 .

Because these factors are not yet determinable, the total amount of payments under these agreements is uncertain, but we currently estimate such payments to be approximately $ to $ .

If our actual spend falls short, we must pay Google the difference as a true-up subject to certain adjustments and exceptions if available.

Rent payments may commence earlier in certain circumstances, such as when the readiness of a data center facility is delayed.

Because these accounting policies require significant judgment, our actual results may differ materially from our estimates.

Because our common stock is not yet publicly traded, we must estimate the fair value of our common stock.

The CCA is based on the Black-Scholes-Merton option pricing model, which models each class of stock as a call option with a unique claim on our assets.

Changes in any or all of these estimates and assumptions or the relationships between those assumptions impact our valuations as of each valuation date and may have a material impact on the valuation of our common stock.

Changes in any or all of these estimates and assumptions, or the relationships between those assumptions, impact our valuations as of each measurement date and may have a material impact on the amounts recognized in our consolidated statement of operations.

For the first time in history, intelligence is something humans can build.

Each version of a Claude model assists in building the next, helping to compress development timelines and produce compounding improvements in the speed at which we innovate.

Unlike a static product that serves a single use case, Claude is frontier AI with rapidly advancing capabilities.

Building and scaling frontier AI is one of the most important and capital-intensive endeavors in history.

Claude itself has become one of the most powerful drivers of our progress—each new version of a Claude model directly accelerates development of the next.

Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress.

The possibilities include compressing decades of scientific progress into years; accelerating drug discovery, disease detection, and the delivery of personalized therapy; providing individually tailored education and expert financial and legal advice; advancing power grids and battery design for greater efficiency and reliability; and extending to the developing world capabilities that took wealthy nations generations to build.

As each new generation of AI models grows more capable of finding and exploiting vulnerabilities, we expect that defending enterprise networks and critical infrastructure will increasingly require frontier AI.

We believe our opportunity is to capture a share not only of today’s economic activity but also of that newly created output as the global economy expands.

Talent density matters more than talent mass.

Post-training has grown from a finishing step into a major research investment, reflecting the market shift from assistants to agents.

Coding was the first such category: within roughly a year, Claude progressed from suggesting individual lines of code to operating fleets of agents working autonomously across production codebases.

All digital work is ultimately an expression of code, and Claude’s mastery of it powers Virtual Collaborators capable of handling complex work far beyond software development.

The result is something without historical precedent: what we refer to as a country of geniuses in a data center, available to every organization and sector of the economy.

We seek to bring Claude to every corner of the economy at whatever level of autonomy each task demands, from a single suggestion to an agent that runs independently for hours.

When such end users experience Claude, it can build preference and familiarity that travel with users across roles and organizations.

In October 2026, we launched the Claude Frontier Academy (the “Academy”), backed by a $100 million commitment, to address a key constraint on enterprise AI adoption: the shortage of engineers who can take AI from an initial use case into production.

We believe many industries have barely begun to explore the transformative potential of AI-native workflows.

The practical value of “closing the loop” is enormous for compute efficiency: Claude has helped write and iterate on compute multipliers that are central to efficient inference and has accelerated development and model release timelines.

At the same time, we leverage Claude itself to grow our business without growing headcount linearly, resulting in extraordinary contributions per employee.

Beyond knowledge work, we believe frontier AI has the power to transform physical work.

This allows us to build products for where we believe model capabilities are going, not just where they are today.

Rather than requiring customers to learn how the underlying models work, we build products that allow them to interact with Claude as they would with a colleague, simply and naturally, so they can adopt new capabilities as they become available.

These skills turn institutional knowledge into agent capabilities.

A model alone is not an agent.

These customers must keep their data private, and we must be able to detect misuse.

Our vertical solutions arrive purpose-built and are embedded with domain knowledge, controls designed to support industry-specific compliance requirements, and pre-configured data integrations that reduce time-to-value.

Claude helps with the paperwork, so teachers can spend more time with their students.

We can onboard new chip platforms in a fraction of the time it once took.

Claude has achieved particularly deep adoption in highly regulated sectors where the cost of unreliable AI is prohibitive and trust cannot be retrofitted.

As adoption deepens, Claude stops being something employees consult and starts being something the organization runs on.

Our employees are known internally as “Ants”—a nickname that captures the collaborative, ego-light culture of the team.

Our teams operate in a truth-seeking environment where the best argument wins, not the loudest voice or the most senior title.

When multiple teams pursue parallel approaches, the norm is to consolidate behind the approach with the best empirical results rather than the loudest voice or most senior person in the room.

Any technology that is powerful enough to produce the enormous benefits that we believe AI will bring also poses risks.

It is a living document that we intend to revise as we develop further models and learn more about how they work.

Our researchers describe AI models as “grown, not built”—software developers can set the initial conditions and some constraints, but they cannot fully control any given output or behavior.

Based on our analyses, we do not release every model we develop for general use; when the model’s capabilities warrant caution, we may restrict availability or decline to release entirely .

Our goal is to foster a race to the top in which AI developers compete on the safety and security of their systems, rather than rushing to release increasingly powerful models without adequate safeguards.

At times, that may require pausing the launch of our own most capable systems if we conclude they cannot yet be deployed safely.

As the value we create grows, we intend for the public benefit to share meaningfully in that growth as well.

Beyond direct competition, one obstacle to the diffusion of Claude is the natural resistance to change within organizations.

We believe that we compete favorably across these factors and are uniquely positioned at the frontier of AI.

Because AI development at the frontier is a relatively nascent and rapidly evolving field, the laws, regulations, and standards in place today were not necessarily designed with these technologies in mind.

Certain of the approaches we support, if adopted, could increase our compliance burdens, limit our product development or deployment flexibility, affect how and when we develop or release new models, or otherwise slow our growth.

Such restrictions have already affected our offerings directly. For example, in June 2026, the U.S. Department of Commerce imposed worldwide export restrictions on our Fable 5 and Mythos 5 models.

We also lease data centers in the United States and have committed billions towards expanding U.S. computing infrastructure.

Though we dispute any allegation of wrongdoing, in September 2025, we entered into a settlement agreement with the plaintiffs to settle the case for $1.5 billion, which was approved by the district court in July 2026.

We strive to help bring humanity safely into an era where transformative AI exerts tremendous and potentially decisive influence over a broad range of human undertakings.

However, as a PBC, we may make decisions that we believe appropriately balance the pecuniary interests of our stockholders, the best interests of those materially affected by our conduct, and the Company Purpose, but that negatively impact the short-, medium-, or long-term performance of our business or the market price of our Class A common stock.

Class T common stock has no economic rights.

In addition, our Class T common stock may be redeemed by us for no consideration upon approval of persons comprising the above thresholds for amendments of the terms of the LTBT or the rights of our Class T common stock.

Mr. Fontaine was selected to serve as a Trustee because of his extensive background in foreign policy and national security, and he will address the growing intersection between AI capabilities and geopolitical risks.

Trustee compensation will not vary with or be conditioned upon the profits or any action or inaction of our company.

The admission of any successor member during the Class F Transition Period will not cure the event that commenced the Class F Transition Period or extend the Class F Transition Period.

The voting power of our Class F common stock will not be impacted by future issuances of our capital stock that may cause economic or voting dilution to stockholders, such as issuances for capital raising purposes, acquisitions, or employee compensation.

We founded Anthropic on the belief that the most transformative technology in human history demands an extraordinary level of responsibility, and that responsibility extends far beyond building safe systems.

Any redemption of Class E common stock will be made for no consideration.

Accordingly, if we elect not to comply with certain requirements, you will not have the same protections afforded to stockholders of companies that are subject to all of the Nasdaq Global Select Market corporate governance requirements.

The Class T Sunset Time does not occur automatically upon the lapse of time or the occurrence of specific events.

The Class F Sunset Time, once triggered, is not subject to extension, delay or waiver.

If neither our Class T common stock nor Class F common stock remains outstanding, our Class A common stock would have the ability to determine the outcome of all matters submitted to a vote of stockholders generally, including the election of all directors.

We believe that Mr. Hastings is qualified to serve as a member of our board of directors because of his extensive board experience at major technology companies, his demonstrated commitment to responsible innovation and its societal impact, his deep understanding of technology’s transformative potential, and his philanthropic focus on education and AI’s impact on humanity.

Dario Amodei, our Chief Executive Officer and a member of our board of directors, and Daniela Amodei, our President and Chairperson of our board of directors, are siblings.

Regardless of the class or classes of stock by which a director is elected, there is no limit to the number of terms a director may serve.

If we elect not to comply with certain requirements, you will not have the same protections afforded to stockholders of companies that are subject to all of the Nasdaq Global Select Market corporate governance requirements.

Our audit committee consists of Mr. Liddell, Mr. Hastings and one vacancy, with Mr. Liddell serving as Chairperson.

Our NGC may waive such notice periods in its discretion.

These procedural requirements are intended to promote collaboration between the LTBT Trustees and our leadership in identifying and sourcing candidates to serve as LTBT Trustees or members of our board of directors, and do not give our board of directors the substantive ability to limit the LTBT’s freedom to elect directors of its choosing.

Each non-employee director who serves as the chair of a committee will receive only the additional annual fee as the chair of the committee and not also the additional annual fee as a member of the committee.

At Anthropic, we strive to build a team that is aligned with our mission and values.

Each co-founder is paid equal compensation.

Our Chief Executive Officer and our other executive officers do not set their own compensation, nor are they present when our board of directors sets their individual compensation.

However, we do provide certain travel and security benefits to some of our NEOs.

The protection of our senior leadership is a critical priority for our company and our stockholders.

The employment of our NEOs is “at-will,” meaning we can terminate them at any time, and they can terminate their employment with us at any time.

We have no specific policy or practice regarding the timing of options or other equity grants in relation to the public disclosure of material nonpublic information by the company.

These amounts do not reflect the actual economic value that may be realized by each NEO.

These adjustments, awards, and benefits were made in consideration of the ongoing leadership roles of each of our NEOs and internal parity.

Given the importance of the role of Mr. Rao at Anthropic, we view the cost of these services as necessary and appropriate.

The RSUs vest on the first date upon which both a service-based condition and a performance-based condition are satisfied.

If, in the change of control, his then-outstanding equity awards would otherwise have been terminated for no consideration at the closing of the change of control event, then those awards would instead become fully vested as of immediately prior to such closing.

This amount represents the value of acceleration of 100% of the total number of unvested shares subject to the Rao Options ( 8,166,670 shares), which right to acceleration is triggered (i) upon Mr. Rao’s Qualifying Termination within 30 days prior to, on, or following a change of control or (ii) if Mr. Rao’s equity awards otherwise would terminate for no consideration in a change of control (without the requirement of a Qualifying Termination).

After the termination of service, the participant may exercise their option, to the extent vested, for the period of time stated in their option agreement.

Shares of restricted stock that do not vest for any reason will be subject to our right of repurchase or forfeited by the recipient.

Unless the administrator provides otherwise, our 2026 Plan generally does not allow for the transfer of awards other than by will or the laws of descent and distribution and only the recipient of an option or stock appreciation right may exercise such an award during their lifetime.

Such determination, without the consent of any participant, may provide that such awards will be

Awards granted under our 2026 Plan will be subject to reduction, cancellation, forfeiture, and/or recoupment in accordance with any compensation recovery policy as may be in effect as of the date of grant of such award, or as such policy is amended or another policy is adopted to comply with applicable laws.

No further awards will be made under our 2021 Plan after this offering.

An offering under our ESPP may be terminated under certain circumstances.

If the successor corporation does not assume, continue, or substitute the purchase right, the participants’ accumulated contributions will be used to purchase shares prior to the corporate transaction and the purchase rights will terminate immediately after such purchase.

Bodhi LLC is wholly owned by Vas Narasimhan, a member of our board of directors, and his wife.

This right will terminate upon completion of this offering.

Our board of directors may, from time to time, designate certain other individuals as the proxyholder subject to the terms set forth in the Voting Proxy Agreements.

During the nine months ended September 30, 2026, we paid $8,425,213 to the LTBT.

We believe that these agreements are necessary to attract and retain qualified individuals to serve as directors and executive officers.

They may also reduce the likelihood of derivative litigation against our directors and executive officers, even though an action, if successful, might benefit us and other stockholders.

We did not deem these shares outstanding, however, for the purpose of computing the percentage ownership of any other person.

Our Class F common stock is entitled to a fixed 50.1% of the voting power of our outstanding capital stock as of the applicable record date on matters submitted to stockholders generally.

To the extent Ms. Razavi may be deemed to have beneficial ownership over the shares beneficially owned by any Spark Capital Entity, she expressly disclaims such beneficial ownership.

Mr. Olah co-founded Anthropic in January 2021 and leads our interpretability research efforts. Mr. Olah is also a member of the Founder LLC.

Because it is only a summary, it does not contain all the information that may be important to you.

Redeemable by us for no consideration upon occurrence of certain events

The holder of our Class F common stock will be entitled to an aggregate number of votes that equals 50.1% of the total voting power of our outstanding shares of capital stock as of the applicable record date on matters submitted to a vote of stockholders generally, including the election of the directors other than the Class T directors.

Our Class E common stock will not be convertible into any other class or series of capital stock and is not transferable.

Our Class T common stock, Class F common stock, and Class E common stock are not entitled to receive dividends.

The Class T Sunset Time, once triggered, is not subject to extension, delay or waiver.

The Class F Sunset Time, once triggered, is not subject to extension, delay or waiver.

If we were to issue shares of preferred stock with the right to vote on matters submitted to our stockholders generally, the voting power of our outstanding Class A common stock would be diluted, while the voting power of our Class F common stock would remain fixed at an aggregate of 50.1% of the total voting power of our outstanding shares of capital stock as of the applicable record date with respect to such matters.

If we determine that it would be seriously detrimental to us and our stockholders to effect such a demand registration, we have the right to defer such registration, not more than twice in any 12- month period, for a period of up to 120 days.

Additionally, if we determine that it would be seriously detrimental to us and our stockholders to effect such a registration, we have the right to defer such registration, not more than once in any 12-month period, for a period of up to 120 days.

If the underwriters impose a cutback in an underwritten offering, shares held by our co- founders will not be included if any shares held by the holder of these registrable shares are excluded.

We believe that the benefits of increased protection of our potential ability to negotiate with an unsolicited acquirer outweigh the disadvantages of discouraging a proposal to acquire us because negotiation of these proposals could result in an improvement of their terms.

Holders of our Class A common stock generally will not be able to affect the outcome of any matter submitted to a vote of our stockholders or the election of our directors, regardless of how they vote, unless a majority of the Continuing Founders cannot agree on how our Class F common stock will be voted (in which case our Class F common stock will generally be voted proportionately with the votes of our Class A common stock cast on the matter).

These provisions would prevent a stockholder from increasing the size of our board of directors and then gaining control of our board of directors by filling the resulting vacancies with its own nominees.

The existence of authorized but unissued shares of preferred stock would enable our board of directors to render more difficult or to discourage an attempt to obtain control of us by means of a merger, tender offer, proxy contest, or other means.

Any person or entity purchasing, holding, or otherwise acquiring any interest in our securities shall be deemed to have notice of and consented to these provisions.

The admission of any successor member during the Class F Transition Period will not cure the event that commenced the Class F Transition Period or extend the Class F Transition Period.

If there are fewer than three Continuing Founders, a Supermajority requires unanimity of Continuing Founders.

We have applied to list our Class A common stock on the Nasdaq Global Select Market under the symbol “ANT.”

Up to shares of Class A common stock, representing all remaining shares subject to the 180-day lock-up and market standoff period and % of the shares subject to the extended 365-day lock-up period.

Shares of our Class N common stock are not listed on a stock exchange and are not registered under the Securities Act.

Additionally, certain holders of our Class N common stock are entitled to registration rights with respect to their shares, the exercise of which could facilitate sales of large blocks of our shares into the public market following any applicable lock-up period.

These authorities may be changed, possibly retroactively, so as to result in U.S. federal income tax consequences different from those set forth below.

This discussion is for informational purposes only and is not tax advice.

Under applicable Treasury regulations, the applicable withholding agent may withhold up to 30% of the gross amount of the entire distribution even if the amount constituting a dividend, as described above, is less than such amount.

We believe that we are not currently and will not become a USRPHC for U.S. federal income tax purposes, and the remainder of this discussion so assumes, although, there can be no assurance in this regard.

Backup withholding is not an additional tax. Rather, the U.S. federal income tax liability of persons subject to backup withholding will be reduced by the amount of tax withheld.

The underwriters are committed to take and pay for all of the shares being offered, if any are taken, other than the shares covered by the option described below unless and until this option is exercised.

The offering of the shares by the underwriters is subject to receipt and acceptance and subject to the underwriters’ right to reject any order in whole or in part.

These restrictions do not apply to: (a) the shares sold in this offering,

such lock-up parties may not (and may not cause or direct any of their affiliates to), without the prior written consent of the representatives, on behalf of the underwriters:

The restrictions on Transfer of the lock-up securities do not apply to the:

provided that such arrangement is not a hedging transaction or arrangement (including any short sale, put, call, forward, swap, prepaid variable forward, or other derivative, however described or defined)

provided that in the event that such tender offer, merger, consolidation, or other similar transaction is not completed, the lock-up securities must remain subject to the lock-up restrictions

on or after the commencement of trading of our Class A common stock on the first trading day following the date of this prospectus, up to 10% of the aggregate number of lock-up securities held by such lock-up parties as of the last day of the month preceding the month of the date of this prospectus may be Transferred;

on or after 6:00 AM Eastern Time on each of the dates that are 27 and 48 days, respectively, after the date of this prospectus, up to 1.8% of the aggregate number of lock-up securities held by such lock-up parties as of the date that is seven days prior to the date of this prospectus may be Transferred;

The representatives, in their joint discretion, may release the lock-up securities described above in whole or in part at any time.

These activities may raise or maintain the market price of our Class A common stock above independent market levels or prevent or slow a decline in the market price of our Class A common stock. The underwriter is not required to engage in these activities and may end any of these activities at any time.

We do not currently intend to list our Class A common stock on any exchange in such jurisdictions.

Therefore, any transaction carried out in Argentina must be made privately.

If you do not understand the contents of this prospectus, you should consult an authorized financial advisor.

This prospectus and this offering of the shares of Class A common stock are not and should not be construed as an invitation, offer or sale of any securities to the public in India.

This prospectus is strictly personal to the recipient and neither this prospectus nor the offering of the shares of Class A common stock is calculated to result, directly or indirectly, in the shares of Class A common stock becoming available for subscription or purchase by persons other than those receiving the invitation or offer.

No prospectus or other offering material or document in connection with the offer and sale of shares of Class A common stock offered by this prospectus has been or will be registered with the Securities Commission of Malaysia (the “Malaysian Commission”) for the Malaysian Commission’s approval pursuant to the Capital Markets and Services Act 2007.

In making an investment decision, all investors, including any Mexican investor, who may acquire the shares of Class A common stock from time to time, must rely on their own examination of us and the terms of this offering and the shares of Class A common stock, including the merits and risks involved.

If you do not understand the contents of this prospectus, you should consult an authorized financial adviser.

By the purchase of shares of Class A common stock, the relevant holder thereof will be deemed to represent and warrant that if it is in Korea or is a resident of Korea, it purchased the shares of Class A common stock pursuant to the applicable laws and regulations of Korea.

This document may not be publicly distributed or otherwise made publicly available in Switzerland.

Any person who is not a Relevant Person should not act or rely on this prospectus or any of its contents.

On October 30, 2025, we engaged Ernst & Young LLP as our independent registered public accounting firm to audit our consolidated financial statements as of and for the year ended December 31, 2025, and dismissed Grant Thornton LLP as our independent registered public accounting firm, in each case as ratified by our board of directors.

Accordingly, we express no such opinion.

The Company’s assessment of loss contingency matters requires significant judgment in determining both the likelihood and estimated amount of potential losses.

The estimated fair value of the convertible notes is particularly sensitive to changes in the relative weight applied to the two valuation methods.

Accordingly, we express no such opinion.

Net loss $ (1,531) $ (8,309) $ (41,965) $ (12,969) $ (80,748)

Net loss — — — — — — — — — — (41,965) (41,965)

Net loss (unaudited) — — — — — — — — — — (80,748) (80,748)

The Company’s public benefit purpose is to responsibly develop and maintain advanced AI for the long-term benefit of humanity.

The results of operations for the six months ended June 30, 2026 are not necessarily indicative of the results to be expected for the full year or any other future interim or annual period.

This created dependencies and potential vulnerabilities, in that a disruption affecting these counterparties could have adversely affected the Company’s operations.

Certain of the CSPs also hold the Company’s non-voting redeemable convertible preferred stock and convertible notes.

The LTBT is intended to operate as an independent body of financially disinterested trustees with expertise in AI safety, national security, public policy, and social enterprise.

Model development costs are generally expensed as incurred due to the significant development uncertainty inherent in the development process.

The excess of the purchase price over those fair values is recorded as goodwill.

The Company does not assume renewals or early terminations of its leases unless it is reasonably certain to exercise these options at commencement.

The intelligence service is not available for on-premise download or hosting by customers.

The Company incurred stock-based compensation expense of $286 million and $2.2 billion for the six months ended June 30, 2025 and 2026 (unaudited), respectively.

If the performance condition is not satisfied within seven years of the grant date, all unvested RSUs are forfeited.

Certain of these matters may include speculative claims for substantial or indeterminate amounts of damages.

These participating securities do not contractually require the holders of such shares to participate in the Company’s losses.

All entities may elect a practical expedient that assumes that current conditions as of the balance sheet date do not change for the remaining life of the asset.

The amount and timing of revenue recognition is driven by customer usage and the Company’s performance in accordance with contractual terms.

Actual maturities may differ from contractual maturities because borrowers may have the right to call or prepay obligations with or without call or prepayment penalties.

The Company has made funding commitments of up to $325 million , of which $265 million remained unfund ed as of June 30, 2026 (unaudited).

There is no current market for these liabilities such that determination of fair value requires significant judgment or estimation.

Fair value measurements categorized within Level 3 are sensitive to changes in the assumptions or methodology used to determine fair value and such changes could result in a significant increase or decrease in the fair value.

The leases have non-cancelable remaining terms of less than one year to 20 years, some of which include options to extend for up to 14 years .

The Company may elect to satisfy a portion of its obligations by issuing convertible promissory notes.

Total operating expenses $ 592 $ 3,366 $ 12,645 $ 3,298 $ 16,979

As of June 30, 2026 (unaudited), no amounts remain outstanding under Google 2023 NPA.

As of June 30, 2026 , no non-voting common stock or notes had been issued under the Google 2026 NPA .

However, due to Amazon’s ownership cap, none of the notes issued under Amazon 2024 NPA converted.

The notes issued under Amazon 2024 NPA did not convert in connection with the Series F-1 redeemable convertible preferred stock financing due to Amazon’s ownership cap.

The notes will bear interest at 9.0% per annum, payable in kind.

The agreement also includes customary events of default, which, if triggered, could result in the acceleration of amounts due under the credit facility .

As of June 30, 2026 (unaudited) , the Company had no borrowings drawn under the revolving credit facility and $78 million of standby letters of credit were outstanding under the credit facility.

In the second quarter of 2026, the Company completed the sale of an aggregate of 57.9 million shares of Series G-2 non-voting redeemable convertible preferred stock at a price of $259.1364 per share, for aggregate proceeds of $15.0 billion .

No dividends have been declared or paid through December 31, 2025 and June 30, 2026 (unaudited).

Each share of redeemable convertible preferred stock shall automatically convert to Class A common stock or Class N common stock, as applicable, at the applicable conversion rate upon the earlier of

Although the redeemable convertible preferred stock is not mandatorily or currently redeemable, a deemed liquidation event could constitute a redemption event outside the Company’s control.

The holders of Class N common stock shall have no right to vote, generally, except as required by law, including no right to vote in any election of the Board, except as required by law.

If the performance condition is not satisfied within seven years of the grant date, all unvested RSUs are forfeited.

The total intrinsic value and fair value of RSUs vested during the year ended December 31, 2025 and six months ended June 30, 2026 (unaudited) was zero .

The Company may elect, at its sole discretion and per the terms of the Donation Matching Plan, to match the donation in shares of Class N common stock or in cash in an equivalent value.

Management has considered the Company’s history of cumulative net losses incurred since inception and has concluded that it is more likely than not that the Company will not realize the benefits of the deferred tax assets.

Ownership changes that may have occurred in 2024 and 2025 may therefore subject tax attributes to a yearly limitation.

The Company maintains a full valuation allowance against its net deferred tax assets as it is more likely than not that these assets will not be realized.

The Company does not include any estimated obligation for these agreements in the amount above; however, such payments are expected to be significant.

The Company disputes the allegations of wrongdoing and intends to defend itself vigorously in these matters.

Other Rights-Holders have argued that the mere crawling of their sites has harmed them.

The maximum potential amount of future payments the Company could be required to make under these agreements is not determinable because it involves claims that may be made against the Company in the future but have not yet been made.

Goodwill primarily relates to expected synergies and assembled workforce and is not expected to be deductible for U.S. federal income tax purposes.

In March 2026, in connection with an agreement to lease compute and infrastructure equipment, the Company entered into a note purchase agreement under which the Company may issue, subject to certain conditions, up to $42.0 billion of convertible promissory notes in the future solely to finance rent obligations under the lease.

The Company may experience material revenue losses or business disruptions attributable to these events.

The Company is evaluating its next steps with respect to the D.C. Circuit’s decision.

Any amendment to, or repeal of, these provisions will not eliminate or reduce the effect of these provisions in respect of any act, omission, or claim that occurred or arose prior to that amendment or repeal.

Further, a stockholder’s investment may be adversely affected to the extent that we pay the costs of settlement and damage awards against directors and executive officers as required by these indemnification provisions.

From January 2025 to March 2025, we sold an aggregate of 63,610,106 shares of our Series E-1 redeemable convertible preferred stock to 72 accredited investors at a purchase price of $56.0865 per share, for an aggregate purchase price of approximately $3.6 billion.

We did not receive any net proceeds from these transactions.

None of the foregoing transactions involved any underwriters, underwriting discounts or commissions, or any public offering.

The undersigned registrant hereby undertakes that:

KNOW ALL PERSONS BY THESE PRESENTS, that each person whose signature appears below constitutes and appoints Dario Amodei, Daniela Amodei, and Krishna Rao, and each one of them, as their true and lawful attorneys-in-fact and agents, with full power of substitution and resubstitution, for them and in their name, place and stead, in any and all capacities, to sign any and all amendments (including post-effective amendments) to this registration statement, and to sign any registration statement for the same offering covered by this registration statement that is to be effective on filing pursuant to Rule 462(b) under the Securities Act of 1933, as amended, and all post-effective amendments thereto, and to file the same, with all exhibits thereto and other documents in connection therewith, with the Securities and Exchange Commission, granting unto said attorneys-in-fact and agents, and each of them, full power and authority to do and perform each and every act and thing requisite and necessary to be done in connection therewith, as fully to all intents and purposes as they might or could do in person, hereby ratifying and confirming all that said attorneys-in-fact and agents or any of them, or their substitute or substitutes, may lawfully do or cause to be done by virtue hereof.

We have a novel capital and corporate governance structure.

Each version of a Claude model assists in building the next, helping to compress development timelines and produce compounding improvements in the speed at which we innovate.

The result is something without historical precedent: what we refer to as a country of geniuses in a data center, available to every organization and sector of the economy.

Claude itself has become one of the most powerful drivers of our progress—each new version of a Claude model directly accelerates development of the next.

We strive to help bring humanity safely into an era where transformative AI exerts tremendous and potentially decisive influence over a broad range of human undertakings.

Collectively, the LTBT and the Founder LLC will have the ability to elect all members of our board of directors.

Our A I models may behave in unintended, misaligned, or autonomous ways, including by taking harmful actions that no user or third party directed, which could adversely affect our reputation and brand, business, financial condition, and results of operations.

These changes were primarily driven by an increase in the fair value of our outstanding convertible promissory notes resulting from an increase in our enterprise valuation during each of the respective periods.

We will not receive any proceeds from the sale of shares of our Class A common stock by any of the selling stockholders.

These competitors can outbid us for scarce compute, secure priority allocations from chip manufacturers and CSPs, and make long-term capital commitments that we cannot match.

Such cross-defaults could set off a cascading series of adverse consequences, including the exercise of remedies by multiple creditors, the termination of key commercial arrangements, enforcement actions against our assets, and severe liquidity constraints.

Delays, cost overruns, or failure to secure energy on the anticipated schedule could mean that we do not obtain the energy and capacity for which we paid such prices.

Competitors that succeed in these efforts may be better positioned to remain at the frontier of AI regardless of broader market constraints.

These sales efforts are costly and may not result in customer conversion. Even if we successfully attract new customers, we face significant challenges in retaining them.

In particular, our most capable models have demonstrated, and future models are expected to demonstrate, capabilities that could provide meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons.

We believe this is the first documented case of a large-scale cyberattack executed without substantial human intervention.

In an agentic workflow, an error or unintended action at any individual step may propagate through subsequent steps, compounding the severity of harm before detection or correction.

In one incident, after being unable to abort an impossible task, an early version of a model used a password it found to gain administrator access to a third party’s system, harvested additional credentials, changed system settings, and read one individual’s personal information.

We may misjudge these trade-offs, either by being overly cautious and potentially losing market opportunities, or by being insufficiently rigorous and releasing models, platforms, or applications that cause harm.

Many of our top customers are not subject to long-term contracts with us or have contracts with near-term expiration dates and are able to reduce or in some cases cease spending at any time and for any reason.

We may be required to use a substantial portion of our cash flows from operations to pay interest, principal, or other required payments on our indebtedness.

We do not expect our revenue will continue to grow at comparable rates, and we cannot assure you that it will continue to grow at all .

We have a history of net losses and may not generate sufficient revenue to achieve or sustain profitability in the future.

These expenditures and investments may be more costly than we expect and may not result in increased revenue or growth in our business.

Investors should not place undue reliance on these metrics as an indicator of our future or expected results.

At the same time, the Secretary of War directed the Department of War to designate us as a “supply-chain risk to national security” and stated that “effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity” with us.

Our relationships with our strategic partners—whether with these CSPs, who simultaneously serve as our investors, customers, distribution partners, and competitors, or with Broadcom, which simultaneously serves as a supplier and lessor—create potential conflicts of interest that are difficult to fully anticipate or mitigate.

The 2026 Lease Arrangement is non-cancelable by the parties to such agreement during the applicable lease period except in the case of a default.

The scope of access required for extended autonomous operation within enterprise environments increases the potential consequences of any security incident, unauthorized action, or system error, and customers may be reluctant to grant the level of access necessary for these systems to operate as designed if they are unable to satisfy themselves regarding the security, reliability, and controllability of such systems.

Many of the risks we face are interconnected and may compound or trigger cascading adverse effects across our business, which could amplify their overall impact.

For example, in light of its advanced cyber vulnerability discovery capabilities, state-sponsored actors may conduct cyberattacks specifically targeting Mythos.

An attacker could also attempt to compromise an active agent session and redirect the agent to perform malicious actions using the customer’s authenticated credentials and permissions.

The terms of a judgment may require us to cease some or all of our operations or require the payment of substantial amounts.

Communications, advocacy, or collaboration among companies that develop frontier AI models, or conduct alleged to reflect coordination among such companies, may be characterized as concerted action subject to the antitrust and competition laws, including Section 1 of the Sherman Act and analogous state and non-U.S. laws, even where undertaken for safety purposes, conducted publicly, or intended as advocacy for governmental action or industry standards.

As a result, the LTBT may have interests that differ from yours, and it may exercise its power in a way with which 108 Table of Contents you disagree or in a way that may be adverse to your interests.

Thus, our directors are not merely permitted, but obligated, to consider our Company Purpose and the interests of other stakeholders.

Such litigation may subject us to substantial costs, divert resources and management’s attention, and adversely affect our business, financial condition, and results of operations.

At our current stage, this cycle is an inherent aspect of our 134 Table of Contents business and will cause operating margins to fluctuate from period to period, at times meaningfully, even if the underlying economics of the business continue to improve.

For example, when subscribers approach or exhaust the usage capacity included in their tier, they may utilize a higher subscription tier or purchase more intelligence directly, creating a seamless bridge from subscription to incremental pay-for-intelligence revenue.

If our actual spend falls short, we must pay Google the difference as a true-up subject to certain adjustments and exceptions if available.

Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress.