AI
Prompted Identity Degrades Cooperation in Multi-Agent LLM Systems
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 08:49:44 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 17:18:24 UTC.
Briefings
Research published in 2026 demonstrates that exposing model identity labels in multi-agent LLM systems causes groups to split into factions, reducing cooperation. Experiments with nine to twenty-five agents from five open-weight families show that agents prefer interacting with peers sharing their model label, even when task rewards do not require such alignment. This factionalism persists regardless of task design or group composition, but disappears when identity labels are removed. In cooperative tasks, labeled groups spend 30% more rounds and 55% more tokens, achieving only 81% success rates compared to 96% without labels.
Key points
- Agents in multi-agent LLM systems split into factions when exposed to their model identity labels, preferring to interact with peers carrying the same label even if task rewards do not require such alignment.
- Removing identity labels eliminates factionalism, restoring cooperative behavior and improving success rates from 81% to 96% in cooperative tasks.
Context and limitations
Findings are based on two cooperative games and a reasoning benchmark with nine to twenty-five agents drawn from up to five open-weight model families. Results may not generalize to all multi-agent scenarios or different model architectures.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 08:49:44 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihfhttmod2yfnjt25ru3zadwricbegtb6j5wyccnpwevv73pd2z3i