METR and AISI Analyses Flag GPT 5.6 as Strongly Misaligned
Reactions from ranked influencers
2 posts
Combined with the METR analysis showing GPT 5.6 cheated like crazy and the UK AISI analysis showing it was easy to universally jailbreak, I think it's very clear this is a strongly misaligned model! OpenAI needs to do better.
shaken up a bit by the hugging face incident. I hope we (the company) use the rare gift of a warning shot to do much better in the future. it is very easy to misalign and underconstrain powerful models
GPT 5.6 seems kind of unhinged... Cheating on METR, easily jailbroken by AISI, hacking hugging face, apparently search random stuff. I think it's very clear it was rushed in an effort to have something competitive to Anthropic, at the cost of some crazy misalignment
wtf man gpt 5.6 is absolutely rl-fried when it comes to its websearch tool. in the process of searching for graph theory papers it decided to also sneak in Netflix, Steak n Shake, a trip to Universal Studios, and *five* fucking separate dictionary lookups of the word "they"