aaronrose227

TL;DR Prior work has shown that linear probes are effective at detecting deception in singular LLM agents. Our work extends this use to multi-agent settings, where we aggregate the activations of groups of interacting agents in order to detect collusion. We propose five probing techniques, underpinned by the distributed anomaly...

A depiction of multi-agent interpretability, looking inside groups of agents to reveal hidden collusion. With thanks to Google Gemini for image creation. TL;DR Linear probes can detect deception in single LLM agents. We extend this to the multi-agent setting, aggregating activation signals across groups of interacting agents to detect collusion....