AI Tools Are Eating Security Hygiene Alive

A researcher found a live Baseten GitHub token exposed in a public repository, used it to access their production GitHub org with admin permissions, and responsibly disclosed it. Baseten patched within hours. The thread notes that the researcher was evaluating Baseten as an inference provider, which is almost comically on-brand: an AI infrastructure company nearly compromised by someone doing due diligence on AI infrastructure.

The pattern here is that the explosion of AI tooling, agents, bots, and CI integrations is dramatically expanding the attack surface for credential leaks. People are wiring together more external services than ever, often quickly, and tokens end up in repos, logs, and config files. The Baseten incident is not unique; it is the visible tip of a much larger iceberg that faster shipping and more integrations keeps growing.

Commenters gave Baseten credit for the fast response, but the underlying worry is structural. The more AI companies brag about developer-friendly APIs and easy integrations, the more opportunities there are for a token to land somewhere it shouldn't.