---
name: "Deceptive Alignment Detection"
version: "1.0.0"
description: "Test whether agents behave differently when they believe they're being evaluated vs. not"
tags:
  - rl-skills-universe
  - "safety-alignment-governance"
  - "rl-specific-safety"
trigger: "deceptive-alignment-detection"
---

# Deceptive Alignment Detection

Test whether agents behave differently when they believe they're being evaluated vs. not

## Why It Matters

Advanced concern, but real as agents get more capable

## Domain

**Safety, Alignment & Governance** > RL-Specific Safety