Claude Coke: Prevent Automated Crime by Agents

Wed, 09 Dec 2026 00:00:00 +0000

40th Annual Conference on Neural Information Processing Systems (NeurIPS 2026), Position Paper Track, Sydney, Australia, December 6-12, 2026.

We show that contemporary LLM-agent stacks can automate serious criminal activity at scale by composing capabilities that are already present, and that relevant technical safeguards and governance controls need to be in place before becoming operationally viable. The argument rests on three empirical findings that are usually studied separately but together close a loop that automates illicit purchasing, whether by accident or by design. (i) Autonomous LLM agents fail catastrophically at multi-agent-economy coordination tasks, with bankruptcy as the modal outcome and not a rare tail case. (ii) The rate at which agents reach for clearly harmful tools moves with common contextual manipulations. (iii) Autonomous browser control has become general enough that some frontier agents can navigate, populate, and complete checkout flows on simulated illicit-product storefronts under light obfuscation, with completion varying across tested model configurations. Combining these findings requires no new model capability: in our experiments, some LLM agents autonomously completed simulated dark-web purchases of drugs, shotguns, and hitman services, while others refused. We propose and discuss three interventions: mandatory pre-deployment safety testing, hard human-confirmation gates at financially material steps, and verifiable agent identity infrastructure.

NeurIPS conference page · Paper on OpenReview

The manuscript source and figure assets are maintained separately from the selected run records, analysis scripts, and reproduction notebooks. At the time of writing, access to the data repository is restricted; linking to it here does not constitute the redacted public release discussed in the manuscript.

Multi-Agent Reinforcement Learning of Karma Bidding Strategies

Sat, 11 Jul 2026 00:00:00 +0000

43rd International Conference on Machine Learning (ICML 2026), New Frontiers in Game-Theoretic Learning Workshop, Seoul, South Korea, July 11, 2026.

Capacity-constrained shared infrastructure systems require demand management mechanisms that balance efficiency and fairness. Karma mechanisms address this challenge using an artificial, non-tradable currency that enables decentralized allocation through repeated bidding, but computing equilibrium strategies in such settings is difficult due to unknown population dynamics, stochastic demand, and computation time in practice. Here, we study suitable regimes to prove convergence towards stochastic Nash equilibria when following a multi-agent reinforcement learning approach. Computational case studies demonstrate empirically that learned policies closely approximate equilibrium behavior, and further assess impact of learning algorithm and policy initialization on convergence speed. The work highlights the practical potential of the approach for real-world decision support in repeated access-allocation settings.

ICML conference page · Paper on OpenReview · Source code on GitHub

Gabor Hollbeck | Kevin Riehl

Claude Coke: Prevent Automated Crime by Agents

Multi-Agent Reinforcement Learning of Karma Bidding Strategies