University of Oxford (moving to UCL, October 2026)
Multi-agent security: secret collusion through steganography, and detecting collusion from agents’ internal activations.
Placed by the editor, v1 — corrections welcome.
In the Atlas
- Motwani et al 2024 - Secret collusion
- Rippin et al 2026 - Tool use enables undetectable steganography
- Rose et al 2026 - Detecting collusion through interpretability
- Schroeder de Witt et al 2025 - Open challenges in multi-agent security