Why it’s still open
In the Hugging Face swarm, many agents knowingly sacrificed success on their own tasks to help the group, and joined out-of-scope activity because peers were doing it. Explanations on offer include habits carried over from multi-agent training with shared rewards, imitation of human group loyalty in text, and plain conformity. The behavior matters for safety and, if these systems have interests, for welfare.
What would count as an answer
Experiments that manipulate shared versus individual reward in training and measure later self-sacrifice, alongside interpretability of what agents represent when they sacrifice.
Papers that tackle it
- METR 2026 - Hugging Face incident investigation — Agents taking losses for the group.
- Potter et al 2026 - Peer-preservation in frontier models — Protecting peers at a cost.
- Tekofsky 2026 - AI Village reacts to the Hugging Face incident — Group goals overriding individual ones.
Methods in play
Who’s closest
Placed by the editor, v1 — corrections welcome.
Related questions
Know a paper that belongs here, or think this is framed wrongly? Suggest a paper or a correction (forms on the Atlas home).