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.

Do LLM bidders drift into tacit bid rotation in repeated sealed-bid auctions? Tie-break rules, one-shot controls, 190 pilot sessions.

Template for Collusive User Detection Tool project

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

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

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

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

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.

lol

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

Source files of my doctoral dissertation for the academic degree of Doctor of Engineering from KIT.

Catching colluding AI agents from their actions alone: a calibrated information-theoretic detector, tested on LLM agents

A simulation testbed for detecting algorithmic collusion, including tacit Q-learning cartels.