Towards embedded evaluations for scheming propensities

Research question

What evidence would support a scheming safety argument?

Design & assumptions

Proposes complementary checks spanning training, direct evaluations, deployment and detection.

Findings

A final-model behavioral test cannot by itself establish all required claims.

Necessary claims do not constitute a sufficient safety case.

Limits & counterevidence

Normative technical proposal rather than a measured result; political advocacy paragraphs excluded.

Source evidence

No structured numerical extraction is available in this reading package.