Definition
A flaw in a multi-tenant platform that lets one customer reach, run or modify the AI agents, models or data belonging to another customer of the same vendor. It typically happens when the platform trusts a session identifier as proof of identity, so a neighbouring tenant can be impersonated without valid credentials.
Why it matters
These flaws are among the most dangerous in cloud AI because a compromise on one customer can spill into every other customer on the same platform — a concentration risk executives should assess when choosing agent vendors.