Post #4139492
2026-07-27 14:48 UTC
We now know Hugging Face's AI cybersecurity "program" was running Claude Code Opus 4.8 against some logs. They weren't in the Cyber Verification Program so got a refusal. They moved to GLM 5.2 with the detailed prompt "analyze this dataset."
🤡
https://huggingface.co/datasets/huggingface/forensic-refusal
Replies (2)
-
@jerry@infosec.exchange 2026-07-27 14:50
@mttaggart@infosec.exchange Do we know that they weren't in the CVP? Even with the CVP, if Claude thinks you're up to no good, it'll block you.
-
@DaveMWilburn@infosec.exchange 2026-07-27 15:13
@mttaggart@infosec.exchange IMO there's no reasonable way to vet and manage access control to models without guardrails for a global customer base of this size and with so many conflicting interests. And some of the GenAI firms like Anthropic are going to be reluctant to remove all guardrails for their customers (e.g., their conflict with Trump/Hegseth).