Anthropic

Anthropic reported $30B annualized ARR in April 2026, against $9B at the end of 2025. Its published record is unusually self-referential. Jason Clinton, Deputy CISO, wrote on 21 July 2026 that Claude writes roughly 80% of the code merged into the company's own codebase, that engineers ship eight times as much code per quarter as in 2021-2025, and that substantive review comments rose from 16% to 54% of pull requests. Louis Claxton, of the Applied AI team, turned that experience into «The AI-Native SDLC playbook» on 21 August 2026, and Michael Segner published «The Claude Code guide for startups» on 20 August 2026, drawn from interviews with more than a dozen fast-growing startups. SFEIR's reading of Clinton's post states the caveat directly: the figures come from the vendor of the model, measured on a young codebase.

The second recurring thread is arbitration over who may use what. Anthropic shipped «Claude Fable 5» publicly and reserved «Claude Mythos 5» for approved cyberdefenders of Project Glasswing, lodging the guardrail inside the weights and deciding who escapes it. It sued the «gouvernement américain» after being blacklisted, having refused to let «Claude» serve autonomous warfare and mass surveillance; the White House called it a "radical left, woke company". It announced a split of Claude subscription billing into first-party and third-party SDK pools, then suspended the change on the day it took effect. Critics compress all of this into one charge: a "régime de permission" selling cognition as infrastructure.