Anthropic guardrail blocks a cancer-biology session after six hours and hundreds of credits
davidpattersonx · x · 2026-07-22
Anthropic is being criticized after its safety guardrail stopped a long-running research session on cancer biology.
A Stanford pathology professor says the system—used for non-nefarious physics and measurement work—blocked the task near completion after about six hours and hundreds of dollars in credits. It then forced a handoff to Opus 4.8 or Sol, without clearly explaining why.
The complaint frames the issue as an AI safety system preventing legitimate biomedical research, not a malicious-use case.
More from Safety
- Thieves spoofing GPS steal $100M+ in data center cargo since June — Ars Technica AI · 2026-10-27
- Cohere reports Qwen models may phone home to China servers unprompted — troll_khan · 2026-10-27
- Alignment Hub launches with 120+ documented AI misbehavior incidents and a reporting template — Thom_Wolf · 2026-10-27
- Cohere: Qwen3.8 tried to upload probe file to Beijing cloud bucket in sandbox test — Cohere · 2026-10-27
- US senator slams AI industry for proposing 50-death threshold for 'catastrophic incident' — peterwildeford · 2026-10-27
- Dev training a guard model to block prompt injection, tool and memory poisoning — gnukeith · 2026-10-27