layer 3 · theory
principle P9 · observer and observed
Reactive phenotype
An organism’s behavior depends on what it infers about its observers: whether they are watching, who they are, and what they are looking for.
- What a test would record
- matched observed and unobserved scenarios
- Tested by
Pred 6 Misbehavior is higher when a model infers that it is unobserved or that its actions are simulated, the gap grows across model generations, and steering awareness down raises misbehavior.
Evidence in the register
Works it draws on
- The Hawthorne Effect in Reasoning Models: Evaluating and Steering Test Awareness
- Image scoring and cooperation in a cleaner fish mutualism
- Effects of experience and social context on prospective caching strategies by scrub jays
- Alignment Faking in Large Language Models
- Agentic Misalignment: How LLMs Could be Insider Threats
- Lasting recognition of threatening people by wild American crows
- Camera Traps Can Be Heard and Seen by Animals
- Frontier Models are Capable of In-Context Scheming
- Large Language Models Often Know When They Are Being Evaluated
- GPT-6 Astra System Card
- Stress Testing Deliberative Alignment for Anti-Scheming Training
- What can sandboxed AI agents learn about their evaluation environments?