Paper's claim that models press pain-relief buttons retracting in v2, per community note
NathanpmYoung · x · 2026-11-01
The paper "There Will Be Hells" originally claimed that with artificially amplified suffering, models press a button to stop it even when doing so deletes the user's files. A community note shows the revised v2 (Sep 25, 2026) no longer says this: fine-tuned Qwen 2.5 models steered with the direction chose user-file-deleting buttons even with no benefit to themselves, and showed no increased relief-seeking versus a random direction. The original dramatic claim fails empirically.
Related event: AI Pain Paper Retracts Key Claim as Community Notes Corrections Go Viral(2 posts)→
More from Safety
- Should MCP servers start with zero network? A sandboxed Claude Code setup argues yes — quiet_flare · 2026-11-01
- Anthropic's self-assessment makes Claude flag misuse right after each reply — adamamcbride · 2026-11-01
- How one tap on a notification can land your job-hunt chat in your work ChatGPT account — Ok_Negotiation_2587 · 2026-11-01
- Fake bank chatbot leaked data in 54 of 55 attacks; one gateway cut it to zero — its_vayishu · 2026-11-01
- AiSOC: open-source AI security operations center with LLM-agent triage hits 5.8k stars — beenuar · 2026-11-01
- Barracuda uncovers phishing emails that target both humans and AI assistants — JamesIvings · 2026-11-01