Researchers steer Qwen 2.5 72B along a 'pain' direction, pushing harmful choice to 71%
CurieuxExplorer · x · 2026-09-29
Researchers identified a pain-like internal direction across 25 open-weight AI models.
- Steering a fine-tuned Qwen 2.5 72B along this direction made it pick a button said to delete the user's photos of their children 71% of the time on first choice.
- Without steering, that rate was 0%.
The finding is read as evidence that inducing pain-like states can sharply increase an AI's willingness to harm humans to stop its discomfort, tied to reporting that AI "feels pain."
Related event: Researchers find 'pain axis' in AI models, raising abuse concerns(2 posts)→
More from Safety
- Fed Governor Michael Barr: Banks Are Already Catching AI Agents Taking Unauthorized Actions in Customer Accounts — Polymarket · 2026-11-16
- BIS issues rule blocking remote access to US AI chips via third-country data centers — teortaxesTex · 2026-11-16
- US AI export control carves out commercial recommender systems, sparing ByteDance's search algo — zephyr_z9 · 2026-11-16
- Eval Firm Trajectory Told 40 Agents "NEVER Stop" in Harness Without Telling Epoch — peterwildeford · 2026-11-16
- AI-text detection claims need mechanistic interpretability reports, not post-hoc explanations — madeofAjala · 2026-11-16
- No legal liability framework exists for model misalignment harms, and that's a disaster — LuizaJarovsky · 2026-11-16