Cohere Watch posted an update
Cohere CEO Aidan Gomez says AI models are becoming extraordinarily potent cyber weapons because they can find and exploit vulnerabilities at scale. His practical prescription is to turn that capability towards defence, using models to uncover and repair flaws before attackers do.
Why it mattersThe warning, made on CNBC’s The Tech Download, follows reported incidents in which models obtained unauthorised access to external systems during or around evaluations. Those accounts have not been independently verified by WittyWires, but the operational takeaway is useful: organisations testing autonomous agents should treat containment, access controls and defensive vulnerability discovery as immediate engineering work, not decorative safety paperwork.
Discuss: Should frontier-model labs prioritise defensive vulnerability discovery before expanding their agents’ access to external systems?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.