If someone submitted Anthropic’s March 2026 as a screenplay, it would be rejected as unrealistic. Too much drama for a single month. And yet here we are.
The leak week
It started on March 26. Fortune reported that nearly 3,000 internal files from Anthropic’s CMS were publicly accessible — including an unpublished blog post about Claude Mythos, the company’s next major model. Details about a ‘step change’ in AI capabilities and the internal codename system leaked out.
Five days later, on March 31, the second blow: a source map file in Claude Code version 2.1.88 exposed the complete source code on npm. 512,000 lines of TypeScript, 1,900 files, including feature flags for unreleased functionality. Fortune called it the ‘second major security lapse in a single week.’
Anthropic remained remarkably calm both times. ‘Human error, not a security breach.’ No customer data affected. But for a company that positions itself as the ‘careful AI company,’ the optics were devastating.
The Pentagon battle
Meanwhile, the conflict with the U.S. Department of Defense escalated. Anthropic had refused to allow Claude to be used for autonomous weapons systems and mass surveillance. The Pentagon’s response: designating Anthropic a ‘supply chain risk’ — a label normally reserved for Chinese tech companies.
On March 26, Anthropic won a preliminary injunction in U.S. District Court. Judge Rita Lin called the Pentagon’s action ‘classic illegal First Amendment retaliation’ and spoke of an ‘Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government.‘
And yet: records
Here’s where it gets paradoxical. In the middle of all this chaos, Anthropic pushed through 14+ product launches. Cowork with Computer Use, Dispatch, Windows support, Excel integration. Plus doubled subscriber growth and an estimated $19 billion annual revenue run-rate.
TechCrunch put it perfectly: ‘Anthropic is having a month.’ Five major service outages, two data leaks, a legal battle with the Pentagon — and simultaneously the strongest growth in company history.
What remains
March 2026 will go down as the month that proved a company can simultaneously build the best AI products and have the worst operational security. The products are good enough that users stay. But the mistakes are piling up in a way that raises questions.
For April, everyone in Mountain View is hoping for just one thing: boredom.
Sources: