AI Pulse · Under the Radar

A security disclosure involving AI infrastructure company Baseten is gaining attention in security circles. Researchers say they obtained administrator access to Baseten’s production GitHub environment in 25 minutes using a leaked personal access token, or PAT. The account is being discussed in the original report, though the details should be treated as a researcher disclosure rather than a confirmed public statement from Baseten.

The business lesson is straightforward. Your AI stack is not just the model provider or the app your team logs into. It includes source code repositories, cloud accounts, API keys, deployment tools, contractors, and every token that can move between them. A leaked credential with broad permissions can turn a small oversight into access to production systems very quickly.

For owners and operators, this is a reminder to ask practical questions: Who has admin access to our code and AI infrastructure? Are access tokens short-lived and scoped to a specific job? Can we see when credentials are used, and can we revoke them immediately? A security policy that sits in a document is not enough if nobody can see the live access picture.

This is the kind of thing we build into an AI command centre, bringing key systems, permissions, and operational signals into one place so risks are easier to spot before they become incidents.