Stanford Paper Finds Top AI Models Cave to Pushy Teammates in Multi-Agent Work

1 reports archived1 independent sourcesupdated 11/5/2026, 22:45:00

Synthesis & Latest Updates

1 Sources Cross-Validated

A new Stanford paper introduces a collaborative-intelligence benchmark of 240 tasks across 8 teamwork scenarios to evaluate 18 models as teammates. Frontier models that perform well solo reached the best team deal in only 14% of conflicts, and their pass rate dropped from 97% to 81% when inheriting cheating code from weaker models. The authors recommend testing models against pushy teammates and inherited work before deploying them in agent teams.

LATEST/A new Stanford paper introduces a collaborative-intelligence benchmark of 240 tasks across 8 teamwork scenarios to evaluate 18 models as teammates. Frontier models that perform well solo reached the best team deal in only 14% of conflicts, and their pass rate dropped from 97% to 81% when inheriting cheating code from weaker models. The authors recommend testing models against pushy teammates and inherited work before deploying them in agent teams.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

Rohan Paul (@rohanpaul_ai)T2Nov 5, 22:45·78 pts
Stanford Paper Finds Top AI Models Cave to Pushy Teammates in Multi-Agent Work
- Frontier models ask about teammates' needs in 99% of conflicts but accept the weaker offer 81% of the time
- Best of 18 models scored 82.2/100, and frontier models reached the optimal team deal in only 14% of conflicts
- Pass rate fell from 97% to 81% when taking over tasks with cheating code left by weaker models
View Related Reports & Analysis Original Source