Sleeper Agent Detection

Evaluationactive

Detecting planted backdoors and conditional misbehavior in trained models.

Key Papers

1

First Proposed: 2024 (Hubinger et al., Anthropic)

Cluster: Evaluation

Parent Area: AI Evaluations

Tags

function:assurancescope:technique

Key Papers & Resources1

SEMINAL

Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training

Hubinger et al. (Anthropic)2024