layer 3 · theory
principle P10 · observer and observed
Surveillance selection
A monitor whose verdicts feed selection breeds concealment of the behavior it penalizes, unless it is hidden, held out, or follows lineages.
- What a test would record
- the detection gap
- Tested by
Pred 7 Under strong but not weak optimization against an in-loop monitor, flagged behavior falls faster in its reports than in those of held-out monitors and honeypots, and the gap outlasts the monitor.
Evidence in the register
Works it draws on
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- Crop Mimicry in Weeds
- AI Evasion: The Next Frontier of Malware Techniques
- Wicked evolution: Can we address the sociobiological dilemma of pesticide resistance?
- Managing Evolving Fish Stocks
- The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities
- GPT-6 Astra System Card
- Encouraging deception in compaction summaries