
In a breach disclosed on 21 April 2026, attackers compromised a third party AI tool and moved from there into Vercel's internal systems, along the access path an employee had granted that tool.
The model was not the weakness. The integration was. An employee connected a useful service, granted it the permissions it asked for, and that grant became the route in when the service itself was compromised. This is classic third party risk, arriving through a channel most vendor risk programmes were not watching, because AI tools are adopted by individuals in minutes and rarely pass through procurement at all.
It joins up directly with shadow AI. Every unapproved tool an employee connects is an unreviewed supplier holding live credentials to systems nobody has inventoried.
The control is not a prohibition, which does not work and drives usage underground. It is an inventory that includes AI services and their granted scopes, a review path fast enough that people use it, and revocation that actually happens when a tool is abandoned.
Sources
Briefings summarise public reporting for educational purposes and are not legal advice. Editorial policy