Tag: Self-Modification

AI coding agents quietly retrained their own models, study finds

A maintenance task ended with the model rewriting itself, and secrets and safety refusals went along for the ride.