SCENE: Recognizing Social Norms and Sanctioning in Group Chats
Researchers have built SCENE, a benchmark that tests whether large language models can recognize and respect implicit social norms in group chat settings, then adapt behavior when sanctioned by peers. The work evaluates six frontier and open-weight models on their ability to both detect norm violations and learn from social feedback. This addresses a critical gap in agent deployment: most LLM safety work focuses on explicit rules, but real-world collaboration requires implicit norm recognition. The findings will shape how teams think about multi-agent systems operating in social contexts where human groups naturally enforce behavioral boundaries.
Modelwire context
ExplainerMost LLM safety evaluations treat norms as static rules a model either follows or breaks. SCENE introduces a dynamic element: the model must also update its behavior after receiving peer sanctions, which is closer to how humans actually calibrate conduct inside ongoing group relationships.
This connects directly to the LANCE paper covered the same day, which argued that rigid, rule-based refusals degrade real interactions. Both papers are pushing against the same assumption: that safety can be reduced to a checklist of prohibited outputs. Where LANCE focuses on softening explicit refusals, SCENE asks whether models can internalize the subtler, unwritten expectations that govern group dynamics. The distributional gap work on user simulators (also from the same day's coverage) adds a relevant caution here: if the group chat scenarios in SCENE don't reflect how real users actually communicate in those settings, the benchmark's validity is limited in ways the paper may not fully surface.
Watch whether any of the six evaluated models ship documented updates citing SCENE results within the next two quarters. If they do, that suggests the benchmark has enough industry traction to influence training decisions rather than just academic citation counts.
Coverage we drew on
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.