collusion

Replication of "Artificial Intelligence, Algorithmic Pricing, and Collusion" by Calvano, Calzolari, Denicolò, Pastorello (2020) in Python, Matlab and Julia.

simulation platform for algorithmic pricing

Q-learning oligopoly simulation (Calvano et al., 2020) — high-fidelity Python reproduction with ≥ 95 % accuracy

Can regulators break algorithmic collusion once it's learned? A Q-learning Bertrand duopoly replication testing four regulatory interventions and their persistence after re-convergence.

Apart AI Collusion Sprint (Oct 2026): does LLM agent collusion survive mid-episode agent replacement?

Template for Collusive User Detection Tool project

BSc Thesis: Investigation of the collusive tendencies of Q-learning and DQN agents in economic environments.

Repo for the publication Channeling Collusion: An LLM Agent Rigs Procurement Auctions by Message, and One Sentence Curbs It

Evidence ladder for LLM collusion monitors: what do monitors treat as convincing evidence of collusion, and does it change when animals are the party harmed? Code, prompts, 288 synthetic transcripts and results from the AI Collusion Research Sprint (Oct 2026).

Where do LLM agents store collusion? Provenance + repeated counterfactual audits show it lives in memory — and sanitised memory transmits it.

On Being Honest in Collusive Places: LLM overseers decide who is colluding from the platform record, not from what players do, so one forged line makes them blame the auditor. Code, games, ratings and paper.

What agent logs can and cannot prove about AI collusion: transfer, permission and arrangement, tested against built look-alikes (Apart AI Collusion Research Sprint 2026)

A study of whether LLM agents shift collusion from chat to bidding when told they are being monitored, comparing chat-based, behavioural, and combined detection methods.

The Undercover Firm: a scripted undercover competitor invites LLM pricing agents to fix prices. Apart Research AI Collusion Sprint, Oct 2026.

The Undercover Firm: a scripted undercover competitor invites LLM pricing agents to fix prices. Apart Research AI Collusion Sprint, Oct 2026.

Do announced sting operations deter collusion between LLM agents? Apart Research AI Collusion Sprint 2026

Testing whether selective memory resets break established collusive behavior in LLM agents.

lol

Mechanistic interpretability of LLM agent collusion in continuous double auctions. Caches residual streams and builds contrast pairs for training probes on collusive intent.