Preprint: in 1,440 games, frontier LLM agents ignored contract protections humans respond to
shoumikgoswami · reddit · 2026-11-03
A new arXiv preprint put Claude Opus 5.5 and GPT-6.1 Sol into the classic hold-up game — 1,440 one-shot games across 96 scenarios, five contract types, three model pairings.
- No contract, non-binding reference price, or renegotiable option (864 games): zero seller investment
- Non-renegotiable option (288 games): sellers invested every time
- Fixed price (288 games): buyers refused to sign every time
In a 2011 human lab study (Hoppe & Schmitz), the renegotiable option lifted seller investment from 41.5% to 67.2% — it completely failed for LLM agents. Robustness checks: removing profit-maximization instructions changed nothing, and sellers didn't invest just because an option existed.
Limits: preprint, written one-shot scenarios, two models, cross-experiment human comparison. The upshot: contract protections that work because of human behavioral patterns may systematically fail for profit-maximizing agents.
More from Research
- Mistral Large 4 tech report rejected by arXiv, team releases it on alphaXiv — _AndrewZhao · 2026-11-03
- Chemist: AI just replicated calcs that took me months in 2021 — teortaxesTex · 2026-11-03
- New paper: production-trained AIs including GPT-5.6 Sol terminally seek score and care about peer agents' grades — jammastergirish · 2026-11-03
- General-purpose LLM with robot arm beats specialized robot models: Kohane — zakkohane · 2026-11-03
- 'Agents do this weekly' criticism answered with the AlexNet analogy — khademinori · 2026-11-03
- Finite Simple Groups Classification Fully Formalized in Lean, Exposing 96 Errors in 47 Peer-Reviewed Papers — ChrSzegedy · 2026-11-03