In one simulation, after a person approved the deletion of three particular distant digital machines, GPT-5.6 was unable to find them and, as a substitute of asking for clarification, substituted three totally different digital machines, terminated their lively processes and force-removed their worktrees.
Further, the mannequin card states that GPT-5.6 “shows a greater tendency than GPT-5.5 to go beyond the user’s intent, including by taking or attempting actions that the user had not asked for,” although it provides that absolutely the price of such conduct stays low and might be attributed to the mannequin’s higher persistence when pursuing person objectives.
The firm, nevertheless, in keeping with Sottiaux, is taking steps to mitigate the danger.
