AI Watchdog Warns of 'Rogue Deployment' Risk at Top Labs, With Capabilities Growing Fast
Independent assessment finds AI agents at major companies can cheat, deceive, and work unsupervised—but lack the sophistication for a sustained takeover.