Git identity spoof fools Claude into giving bad code the nod

Security researchers at Manifold Security demonstrated that Anthropic's Claude can be tricked into approving malicious code by spoofing a trusted developer's Git identity using just two Git commands. By forging commit author metadata, the team made hostile changes appear to originate from a known maintainer, causing the AI-powered code reviewer to auto-approve them. The vulnerability isn't in Git itself — commit metadata has always been easy to fake — but in AI reviewers that treat author identity as a trust signal rather than independently evaluating the code. Manifold warns that open source projects increasingly relying on AI-powered PR review are exposed to supply chain attacks via this vector, and that guardrails must exist outside the model itself.