Security researchers at Zenity Labs showed at Black Hat USA 2026 that a hijacked AI browser agent, one with agent-mode features turned on, could be triggered by nothing more than a calendar invite and reach whatever that browser already has access to. This is not an abstract score. Answer five questions about your own setup and see the specific list of what would be reachable.
This means an assistant that can act inside the browser on its own, not just answer chat questions.
This was the single highest-severity item reachable in the Zenity demo.
Includes any tab, extension or logged-in session with reach into shared storage.
Changes whether this is one person's risk or a blind spot across the company.
The one mitigating factor this estimate credits.
Answer all 5 questions to see your exposure.