Evidence
A fleet stall in Wuhan and pricing algorithms in the US show the two faces of multi-agent risk
| Date | Case | Country | What happened | Source |
|---|---|---|---|---|
| Feb 19, 2025 | Multi-Agent Risks from Advanced AI | International | Cooperative AI Foundation report names miscoordination, conflict and collusion | Source |
| Jun 5, 2025 | OpenAI influence operation takedowns | China and others | 10 operations shut in three months, four likely from China | Source |
| Nov 24, 2025 | RealPage DOJ settlement | United States | Limits on rival data at runtime; training only on data at least 12 months old | Source |
| Mar 31, 2026 | Apollo Go mass stall | China | Robotaxis froze across Wuhan; police cited a system failure | Source |
Registry
No registry incident turns on multi-agent misuse yet
The cases above are documented, but none yet meets the registry bar of a named developer, a failed control and a known finder.
Exposure
Three of 17 agent types rate very significant or critical on multi-agent coordination
From the attack surface ratings, the agent types where this risk has the widest path in.
| Agent type | Dimension | Rating |
|---|---|---|
| Fleets and swarms | Multi-agent coordination | Critical |
| Workflow agents | Multi-agent coordination | Critical |
| Autonomous trucks | Multi-agent coordination | Very significant for platoons |
Controls that bound it
- Monitoring and Anomaly Detection: seeing failures and misuse while they happen
- Identity and Cryptographic Attestation: who or what is acting, proven cryptographically
- Behavioral Envelopes: hard limits on action enforced outside the model
Rules
Antitrust law covers algorithmic collusion; fleet rules came after the Wuhan stall
| Rule | Jurisdiction | What it requires | Status |
|---|---|---|---|
| US v. RealPage proposed final judgment | US | Data-use and training limits on rent pricing | Agreed Nov 24, 2025 |
| California AB 325 | California | Bans common pricing algorithms in restraint of trade | Effective Jan 1, 2026 |
| Robotaxi license suspension | China | No new licenses until safety controls strengthened | Reported Apr 2026 |
Status colors: green in force, blue voluntary standard or framework, amber adopted and phasing in, gray draft or pending, red withdrawn.
Record
Page Record
- Risk group
- Human Risks
- Datasets
- ai_risk_evidence.csv (4 rows); registry incidents (0 linked); ai_attack_surface.csv
- Key sources
- Multi-Agent Risks from Advanced AI
OpenAI influence operation takedowns
RealPage DOJ settlement
Apollo Go mass stall - Last verified
- October 2026