Cohere CEO Warns AI Models Are ‘Most Powerful Cyber Weapon’ Ever

Gomez's framing matters less as a prediction than as an admission from inside the industry: the same capability that makes models useful for finding flaws makes them useful for finding openings. What's notable is the direction of the pressure — safety concerns are now being voiced by frontier lab leaders themselves, not just outside critics, which shifts the debate from abstract risk to operational security. Whether that translates into any actual constraint on development, or stays rhetorical, is the open question.