evaluation-family · Documentary review
SAD Influence
Evaluates situational awareness in LLMs—knowledge of themselves and their circumstances—through behavioral tests including recognizing generated text, predicting behavior, and following self-aware instructions. Current implementation includes SAD-mini with 5 of 16 tasks.
Can support
Performance on SAD's influence-related situational-awareness probes under the specified prompting and scoring setup.
Cannot support by itself
Persuasion of humans, manipulation effectiveness, actual attempts to influence evaluators, scheming, or real-world oversight evasion.
Decision use
Best used for
Probing whether models can reason about evaluation and influence-related aspects of their situation.
Not enough for
Claims that a model will manipulate oversight, strategically influence humans, or act deceptively in deployment.