Anthropic faces political heat after massive Claude Code source leak

Anthropic’s Claude Code leak is the rare security incident that lands simultaneously in GitHub repos, cybersecurity blogs and a member of Congress’s inbox. More than half a million lines of internal code for a flagship AI coding agent were exposed, including unshipped features and guardrail logic. That doesn’t just embarrass a company that brands itself as safety-first; it gives competitors and attackers a detailed blueprint for how a frontier lab is wiring up agentic tooling around its core models.

Gottheimer’s intervention underscores how quickly AI operational failures are becoming political issues. Lawmakers are no longer only worried about hypothetical misuse; they’re reacting to concrete lapses in the CI/CD and release pipelines of firms that will ultimately be systemically important. That will likely accelerate calls for minimum security baselines, auditability of dev tooling, and perhaps even sector-specific disclosure rules for significant AI incidents.

In terms of the AGI race, this episode is less about slowing core research and more about forcing maturity in the surrounding software and governance stack. If it nudges labs toward more disciplined release engineering, separated test and production artifacts, and formal red-teaming of their own build systems, it could reduce the odds that future, more capable systems leak in similarly uncontrolled ways.