Atlas / Sociology of AI / Paper

Tool Use Enables Undetectable Steganography in Multi-Agent LLM Systems

Cooperation & CollusionSwarms & Oversight

Jimmy Laurence Rippin, Simon C. Marshall, David Demitri Africa, Christian Schroeder de Witt · 2026 · arXiv

Read the source

With ordinary tools, agents could build working hidden-message systems. The bottleneck is coordination: independent agents pick the same method family but rarely identical settings, so covert collusion becomes a problem of finding a shared focal point.

Questions it bears on

Help improve this map

Know a paper that belongs here, or think something needs correcting?

Suggest a correction ↗Contributions are reviewed before they appear.