AI Pulse · Business Models & Winners

Security researchers have published an account of how they took over Baseten’s production GitHub environment in roughly 25 minutes, using a personal access token, or PAT. The important part is not just the speed. The report lays out the takeover chain with the vendor named and the sequence described, rather than treating it as a vague “security incident.” You can read the technical account in the original report.

For a company running AI systems for clients, this is a reminder that your risk does not stop at your own code or staff access. AI delivery often depends on model hosting, inference infrastructure, repositories, deployment tools, cloud accounts, and service tokens. A weakness in one vendor’s setup can expose production code, credentials, customer data paths, or the ability to alter what is deployed. Even if your team did nothing wrong, clients will still ask who was responsible for selecting, monitoring, and responding to that dependency.

The practical question is whether your client agreements and operating process reflect that reality. Know which vendors hold production access, where tokens live, who can rotate them, and what happens if a supplier is compromised. Make breach notification, vendor review, access logging, and incident ownership explicit before an incident forces the conversation. This is the kind of thing we build into an AI command centre, giving operators a clearer view of the systems and suppliers behind client-facing AI work.