Not a startup - a government administration running an AI agent over its entire codebase. Alberta did it. And announced it publicly, with numbers on the table.
Fact
The facts: Alberta's Ministry of Technology and Innovation used Claude Code (with the Opus and Sonnet models) to review 466 million lines of government code in about 20 hours, finding and fixing security vulnerabilities across various systems. The project has been running since 2025 and now includes standing agents for continuous security review. Primary source: Anthropic (anthropic.com).
The interesting part isn't the speed, it's the framework around it. A public institution with sensitive code runs an agent, but doesn't give it free rein - it digs and proposes, the human decides what goes in. And it runs continuously, not for a single audit.
The same move is within a small team's reach too: run an agent through the code, you stay at the gate on the fixes. At scale, regular machine review beats manual - that's where the gain is, not in the show around the number 466 million.
Official primary sources
Anthropic - Government of Alberta cybersecurityThe visual is generated code-art, no third-party images.