How AWS averted an AI coding supply chain disaster

AWS discovered that a threat actor extracted an overly permissive GitHub token from a CodeBuild environment and used it to commit malicious code to the open-source Amazon Q Developer VS Code extension; the malicious code was distributed in a release but failed to run due to a syntax error. AWS revoked the compromised credentials, removed the malicious code, released a patched extension, and implemented additional build protections; the report draws lessons about AI-agent fragility, prompt-injection amplification, and the need to treat developer extensions and agents as privileged software.