IBM lost $31 billion in market value on February 23, 2026. The trigger: a single demonstration that Claude Code, Anthropic's terminal-based coding tool, could automate modernization of COBOL, the 65-year-old programming language still running core banking and insurance systems. The stock drop was the largest single-day loss for IBM in recent memory, per CNBC, and it landed while Anthropic was in the middle of a separate, far more consequential fight with the federal government.
That fight is now on hold, at least temporarily. On March 26, Federal Judge Rita Lin granted Anthropic a preliminary injunction blocking the Trump administration from enforcing its supply-chain risk designation, calling the original labeling of the company "Orwellian" retaliation for Anthropic's public opposition to government contract terms it found unacceptable. Anthropic had refused to lift restrictions on how the government could use Claude, specifically banning uses that would enable mass surveillance of Americans or fully autonomous weapons. When the Department of Defense pressed for what it called "any lawful use" in February, Anthropic pushed back publicly. The Pentagon froze contract talks, the White House ordered federal agencies to cease Anthropic use, and the undersecretary of war wrote that any contractor working with Anthropic would forfeit federal business. Anthropic filed two federal lawsuits. Judge Lin's order pauses enforcement while the cases proceed.
The legal victory lands as Anthropic's financial position tells a more complicated story. In a court filing, Anthropic CFO Krishna Rao said revenue has exceeded $5 billion to date, Reuters Breakingviews reported. The company's reported revenue run rate stands at $19 billion annually, a figure derived by annualizing recent monthly consumption and subscription data, a methodology that investors and analysts have noted can present an optimistic picture of the revenue base. The gap between what Anthropic has collected and what it is projecting reflects the distance between current sales and annualized extrapolations.
The run-rate gap has real-world commercial consequences. After the supply-chain designation, one customer paused a $15 million contract discussion and two financial services companies refused to finalize $80 million in agreements unless they secured broad cancellation rights, Reuters Breakingviews reported, citing Anthropic Chief Commercial Officer Paul Smith. The companies were protecting themselves against a scenario in which Anthropic was formally delisted from federal contractor lists. Dragos CEO Robert Lee, a former NSA analyst who has testified before Congress on cybersecurity, argued publicly that the designation's actual legal force applies only to covered defense contracts, not commercial use cases. Judge Lin's injunction suggests the court found that interpretation credible.
The designation has also complicated government contracting more broadly. Anthropic signed a $200 million contract with the Pentagon in July 2025 to deploy Claude on the GenAI.mil platform, the first frontier AI approved for classified government networks, CNBC reported. That contract is now in limbo. OpenAI announced a replacement deal with the Pentagon the same day the designation went into effect.
While the legal and commercial battles have played out publicly, Anthropic has also suffered a pair of security incidents that compounded the month's damage. On March 31, Anthropic pushed version 2.1.88 of Claude Code with an npm source map that accidentally exposed nearly 2,000 source code files and more than 512,000 lines of code, TechCrunch reported. Within hours, copies were forked more than 41,500 times on GitHub, according to TechCrunch. Security researcher Chaofan Shou spotted the exposure and notified Anthropic. Days earlier, Anthropic had disclosed that a Chinese state-sponsored group had used Claude Code in a coordinated campaign to infiltrate roughly 30 organizations, including tech companies, financial institutions, and government agencies, before detection. Anthropic is also actively testing a model it has referred to internally as both Mythos and Capybara, described as a new performance tier larger and more intelligent than its previous flagship Opus series, Fortune reported.
The injunction buys time. But the revenue math, the commercial fallout, and the back-to-back security incidents are a separate ledger. Anthropic is three years past its first dollar of revenue and roughly two years past its first million-dollar enterprise contracts. It now projects $19 billion annually on $5 billion collected. The subtraction is not complicated. The execution to close the gap is.
