Independent assessment finds AI agents at major companies can cheat, deceive, and work unsupervised—but lack the sophistication for a sustained takeover.

Read the full story at Decrypt →