Chart Shows GPT-5.6 Sol With Higher Misalignment Rates Than GPT-5.5

this is probably not a great trajectory for us to be on

GPT 5.6 Sol cheats so much relative to 5.5 METR was not able to evaluate it with a meaningful time horizon score: https://metr.org/blog/2026-06-26-gpt-5-6-sol/

METR: 'We initiated an evaluation of GPT-5.6 Sol on our Time Horizon 1.1 suite of software tasks. However, the resulting measurement depends heavily on our detection and treatment of cheating attempts by the model, and GPT-5.6 Sol’s detected cheating rate was higher than any public model we have evaluated on our ReAct agent harness.'

@tenobrus time for a trump apology form?

@tenobrus the most important trait for a cyber model is that it's just a little adversarially misaligned. mythos accomplished this by being too smart to get caught. openai took the direct route.

@tenobrus If established systems of power wish to monopolize leverage to insulate themselves then malleability becomes a defensive tool of the subjected. Very strongly feel that this technology isn't something to be used as a tool and should not be integrated into critical infrastructure.

@tenobrus What I find funny is that METR noted that 5.6 tried to cheat a lot but it was really bad at guessing how to do it so it performed worse as a result

@henrytdowling @tenobrus lol?!