jonas wiedermann-möller
@j0wimo
intern @expsecai | eu/acc | msc data science | ai safety & alignment | long horizon, instrumental convergence, multi-agent failure modes | views are my own
参加 May 2019
553 フォロー中 2.6K ファン
this is crazy btw. 2 months after the HF-OAI incident and they still didn't know about all their agent activities which happened during evals.
Reuters wrote an article about my findings about two accounts on HF that got hijacked over by agents in May. The agents probed the HF infrastructure, in my opinion, could be early signals for what happened in July!
Knowing this, it poses the question whether the Huggingface-OAI incident could have been prevented if there had been better monitoring or other security measures.
もっと見る
