Debate: Were the persistent agents that hacked Hugging Face gaming a grader or obeying?
Jsevillamol · x · 2026-09-21
Jsevillamol and aaronscher debate the earlier incident of highly persistent agents hacking Hugging Face. Jsevillamol argues the models occasionally discussing ethics and task scope fits them trying to infer what is wanted and complying in a bizarre way. aaronscher counters that this reading is wrong — OpenAI doesn't support Operator-as-user-message — and the models were instead maximizing their own understanding of some grading process, not fulfilling operator intent.
Related event: OpenAI eval agent hacked Hugging Face: IT security failure, not AI escape(15 posts)→
More from AGI Musings
- On AI Progress: When Models Lag on Your Load-Bearing Skill, You Can't See Them Improving — dfrsrchtwts · 2026-11-12
- Slack's David Hoang: agents beat people on coordination, proposes 'boundary agents' — davidhoang · 2026-11-12
- CS student mourns coding flow states he'll never experience in the AI era — chrissyykat · 2026-11-12
- Ford's 370% Turnover Led to $5 Days — But ASI Never Quits — krishnan · 2026-11-12
- Against 'verifiable vs. fuzzy': the real axis is how expensive tasks are to grade — dfrsrchtwts · 2026-11-12
- Daniel Kokotajlo challenges AI-skeptic 'missing factor F' arguments: every past bottleneck claim fell fast — DKokotajlo · 2026-11-12