Are we blindly giving AI access to everything?

Users are connecting AI tools without understanding the security implications.

In this week's Intel Chat, Chris Luft and Matt Bromiley discuss a security breach at Vercel that originated from a compromised third-party AI tool used by one of its employees. The attacker gained control of the employee's Google Workspace account, which provided access to Vercel's internal environment.

Matt addresses the larger concern: developers with privileged accounts are clicking yes on everything, linking tools together and granting permissions without understanding how things chain together. Adversaries, however, understand exactly how to exploit those connections.

The solution goes beyond fixing over-permissioned MCPs. Organizations need user education around the permissions people are granting to AI tools and third-party services.

The episode also covers North Korea stealing $290M in cryptocurrency from Kelp DAO, DDoS attacks on Mastodon and BlueSky, and three ransomware negotiators sentenced for conspiring with the Black Cat group.