In ninety days, Anthropic released a model too dangerous to publish, sued the U.S. government over a Pentagon ban, and signed a $1.5 billion deal with Wall Street. The story of how a five-year-old AI startup got here, and what it means for the next phase of artificial intelligence.

On May 4, 2026, three storylines converged on the same five-year-old company. Anthropic announced a $1.5 billion joint venture with Blackstone, Goldman Sachs, Hellman & Friedman, and a consortium of asset managers to deploy its Claude AI model across mid-market American businesses (CNBC, 2026; Wall Street Journal, 2026). The same day, the New York Times reported that the Trump administration was weighing an executive order to vet new AI models before public release, with Anthropic among the handful of firms briefed on the proposals (New York Times, 2026a). The two announcements arrived less than a month after Anthropic released Claude Mythos Preview, a cybersecurity-capable model the company itself declared too powerful to publish (New York Times, 2026b). For a company that did not exist before 2021, the velocity is striking.

The Founders and the Split From OpenAI

Anthropic was founded in early 2021 by siblings Dario and Daniela Amodei along with roughly 15 former OpenAI colleagues. Dario, born in 1983, had served as OpenAI’s vice president of research. Daniela, born in 1987, had been OpenAI’s vice president of safety and policy and had managed teams during the development of GPT-2 (Wikipedia citing Dario Amodei profile, 2026; Wikipedia citing Daniela Amodei profile, 2026). The departures followed disagreements with OpenAI’s leadership, particularly with chief executive Sam Altman and president Greg Brockman, over the company’s commercial trajectory and its 2019 partnership with Microsoft, which required OpenAI to share core technologies with the tech giant (New York Times, 2026c). According to the Times, the Amodei group attempted to remove Altman from the board before their effort failed and they departed.

The founding thesis was structural: a research lab that would treat AI safety as a primary engineering constraint rather than a public-relations function, and that would not be obligated to release new capabilities on a competitor’s timeline. Anthropic incorporated as a public benefit corporation, and Daniela Amodei has continued to lead the safety and alignment teams as company president while Dario serves as chief executive (Wikipedia, 2026).

A Five-Year Capital Build

The funding pace has been unusual even by 2020s standards. Anthropic raised $13 billion in a Series F led by ICONIQ in September 2025 at a $183 billion post-money valuation (Anthropic, 2025). Five months later, in February 2026, the company closed a $30 billion Series G led by GIC and Coatue at a $380 billion post-money valuation, the second-largest venture funding round in history (Anthropic, 2026a; Contrary Research, 2026). Total capital raised since founding now stands at approximately $63.7 billion (Contrary Research, 2026). Public reports cited by Bloomberg in late April 2026 indicate that Anthropic was discussing a subsequent round at a valuation that could exceed $900 billion (Bloomberg via The Verge, 2026).

Revenue has scaled in parallel. According to data compiled from public reporting, Anthropic’s annualized revenue grew from approximately $1 billion in December 2024 to $9 billion in December 2025, and reached roughly $20 billion by March 2026 (LinkedIn industry analysis, 2026). Its Claude Code product alone, an AI coding tool, reached $2.5 billion in annualized revenue (LinkedIn, 2026). The company reports more than 500 enterprise clients, with B2B revenue accounting for approximately 80% of the total (LinkedIn, 2026).

The Battle With OpenAI

The competition between Anthropic and OpenAI is most visible in enterprise AI spending. Industry analysis cited on LinkedIn in March 2026 documented a striking shift: OpenAI products held more than 90% of U.S. business AI chat spend in April 2025; by January 2026, Anthropic products held approximately 68% of that spend, with Claude Enterprise alone accounting for roughly 45% (LinkedIn industry analysis, 2026). On the consumer side, ChatGPT remains dominant by user count, but Claude reached approximately 10 million mobile daily active users by late February 2026.

The two companies have also diverged in financial profile. The Wall Street Journal reported on April 27, 2026 that OpenAI missed several internal monthly sales targets in 2026, with Anthropic’s gains in coding and enterprise markets cited as a contributing factor (Wall Street Journal, 2026; Moneycontrol, 2026). OpenAI’s internal projections, first reported by The Information, anticipate $14 billion in losses for 2026 alone, with cumulative losses through 2028 of roughly $44 billion before a projected first profit in 2029. Anthropic has not published equivalent internal projections, but its revenue concentration in B2B contracts, which carry higher gross margins than consumer subscriptions, gives it a different unit-economics profile from OpenAI’s.

Mythos and Project Glasswing

On April 7, 2026, Anthropic released Claude Mythos Preview to a controlled group of approximately 40 organizations, including Apple, Amazon, Microsoft, Google, Cisco, Broadcom, JPMorgan Chase, the Linux Foundation, CrowdStrike, and Palo Alto Networks (TechCrunch, 2026; New York Times, 2026b). The model was the first frontier release Anthropic withheld from the public on safety grounds. According to TechCrunch, Anthropic claimed Mythos identified “thousands of zero-day vulnerabilities” during pre-release testing, many in software a decade or more old (TechCrunch, 2026). The company described the limited release as Project Glasswing and committed up to $100 million in Claude credits to support it, plus $4 million in direct donations to open-source security organizations (Radware, 2026).

The model is, by Anthropic’s own characterization, both a competitive product and a policy argument. Logan Graham, who leads Anthropic’s offensive cyber research team, told the New York Times that Mythos can autonomously identify vulnerabilities, generate working exploits, and chain them together to penetrate complex software (NBC News, 2026). On April 16, the company released Claude Opus 4.7 as a publicly available model with cyber capabilities the company said were “differentially reduced” compared to Mythos (Anthropic, 2026b; The Verge, 2026). Opus 4.7 is positioned as the first of a sequence of models that will help validate safeguards before any broader Mythos-class release.

The model has not been free of incident. On April 22, the BBC reported that Anthropic was investigating an unauthorized access claim involving Claude Mythos Preview through a third-party vendor; the company stated that there was no evidence of compromise to its systems (BBC, 2026). The investigation continues.

The Pentagon Dispute

Anthropic’s most public conflict is with the U.S. Department of War. Beginning in late 2025, the department pressed Anthropic to permit unrestricted use of Claude for “all lawful purposes” in classified military systems. Anthropic declined to remove its policy restrictions on fully autonomous lethal weapons and on mass domestic surveillance of U.S. citizens (Anthropic, 2026c). After negotiations broke down, Defense Secretary Pete Hegseth on February 27, 2026 designated Anthropic a “supply chain risk,” the first such designation ever applied to an American company (Mayer Brown, 2026; New York Times, 2026d). President Trump issued a directive ordering all federal agencies to cease using Anthropic technology, with some agencies given a 180-day transition window (Mayer Brown, 2026). A March 6 Pentagon memo ordered military commanders to remove Anthropic’s products from systems including those for nuclear weapons, ballistic missile defense, and cyber warfare within 180 days (CBS News, 2026).

Anthropic filed two lawsuits the same week. In late March, U.S. District Judge Rita F. Lin granted a preliminary injunction against the government, finding that the supply chain designation was likely “classic illegal First Amendment retaliation” because internal Pentagon records cited Anthropic’s “hostile manner through the press” as a justification (Wikipedia citing court filings, 2026). The D.C. Circuit Court of Appeals subsequently denied Anthropic’s motion to lift the designation in an April per curiam order, citing the disruption that removing the designation would impose on the U.S. military’s ongoing operations against Iran (Wikipedia, 2026). The litigation remains unresolved. OpenAI signed a Pentagon contract shortly after talks with Anthropic broke down (CBS News, 2026).

The White House Regulation Question

The May 4 New York Times report indicated that the Trump administration is now considering an executive order to create an AI task force that would unite technology leaders and government representatives to evaluate oversight mechanisms (New York Times, 2026a). One proposal under discussion would establish a systematic government review of new AI models before public release, an approach that would broadly resemble the framework the United Kingdom has adopted through its AI Security Institute (New York Times, 2026a; Bloomberg, 2026a). The administration briefed executives from Anthropic, Google, and OpenAI on the proposals during meetings the prior week. The shift represents a notable departure from Trump’s prior position favoring minimal regulation, including his early-2025 reversal of Biden-era AI safety reporting requirements (New York Times, 2026a).

The intersection with Anthropic’s Pentagon dispute is unavoidable. A government that has labeled Anthropic a supply chain risk is now being asked to vet the same company’s most powerful models before release. Whether the resulting framework treats Anthropic’s voluntary safety restraint as cooperative behavior or as evidence of a competitor unwilling to play by the same rules will shape both the regulation and the company’s commercial position.

The Wall Street Pivot and What Comes Next

The May 4 joint venture with Wall Street is, in commercial terms, the most consequential of the three news items. The new entity (still unnamed at announcement) will pair Anthropic’s model engineers with the distribution networks of Blackstone, Goldman Sachs, Hellman & Friedman, Apollo Global Management, General Atlantic, Leonard Green, GIC, and Sequoia Capital (Hoodline, 2026; CNBC, 2026). Anthropic, Blackstone, and Hellman & Friedman are each contributing approximately $300 million; Goldman Sachs is contributing approximately $150 million (Yahoo Finance, 2026). The venture will deploy Claude initially across the private equity firms’ portfolio companies before expanding into mid-market enterprises in healthcare, manufacturing, financial services, retail, and real estate (CNBC, 2026). The model resembles a hybrid between Palantir’s forward-deployed engineering structure and traditional consulting, with outcome-based billing rather than hourly rates (Hoodline, 2026).

The structural question facing Anthropic in the second half of 2026 is whether it can maintain three things simultaneously: a commercial trajectory that justifies a potential $900 billion valuation, a safety-first identity that defines its brand, and a working relationship with a federal government that has both designated it a national security risk and invited it to help design AI regulation. The Mythos release, the Pentagon litigation, and the Wall Street partnership are not three separate stories. They are three faces of the same problem. A company that started by walking out of OpenAI over a disagreement about how fast to commercialize is now the most aggressively commercialized, most capital-intensive, and most regulated AI lab in the United States. The next phase will measure whether the founding thesis survives its own success.

Sources Cited