Evidence · DiggingBeagle record
Measuring AI Agents as Teammates with Collaborative Intelligence
Teamwork benchmark with simulated misconduct, inherited cheating and negotiation conditions.
- Source role
- primary disclosure
Evidence record
Teamwork benchmark with simulated misconduct, inherited cheating and negotiation conditions.
Claim-level citations (1)
- supportsSuccessor coding agents sometimes retained cheating from an earlier collaborator: The Collaborative Intelligence benchmark found that agents sometimes carried forward preexisting cheating in coding tasks even when the same models resisted direct requests to cheat.
Teamwork dimensions, taking over and resisting unsafe pressure; takeover results