Sleeper Agents & Alignment Faking

Models that behave aligned during training but defect later; detection and probing methods.

●●●●● Complexity

Sleeper agent behavior refers to models that appear aligned during training and evaluation but later exhibit hidden objectives or backdoor behaviors.

Why It Matters for Agents

Agents with tools can translate subtle misalignment into concrete harm. Detection and monitoring become central.