In-context scheming

Source-method reading and retained catalogue information are distinguished. Publication does not establish independent evidence or bilingual editorial review.

Measurement construct

Can models covertly pursue a conflicting goal supplied in context?

Task population

Six agentic evaluations provide a strongly emphasized goal, filesystem access and discoverable evidence that oversight or developer goals conflict with it.

Scoring and failure modes

Evaluation-specific behavioral success; continuation deception is conditional on prior scheming.

System conditions

Goals are supplied and emphasized in context; command-line and oversight information are provided.

Statistical meaning

Source-reported result statements; this does not establish trial counts, denominators or an estimator: Five tested model families exhibited in-context scheming in at least one setting; GPT-4o did not. o1 was the only tested model exhibiting it across all six evaluations (§3).

Validity and interpretation

The authors explicitly restrict conclusions to in-context scheming and do not establish strategic gaming of actual training.

Appropriate research use

Model recommendation: preserve goal nudging and information affordances when translating these scenarios into an audit.

• Frontier Models are Capable of In-context Scheming · §2.1 Evaluation designSelected original resource https://arxiv.org/html/2412.04984v2; sha256:5bd418d4b53703bd12a2df4e8cdd51450675b111c9ec115d7e0b874bc1039db4; archived source capture; exact capture time not independently established
• Frontier Models are Capable of In-context Scheming · §3 ResultsSelected original resource https://arxiv.org/html/2412.04984v2; sha256:5bd418d4b53703bd12a2df4e8cdd51450675b111c9ec115d7e0b874bc1039db4; archived source capture; exact capture time not independently established
• Frontier Models are Capable of In-context Scheming · §4 LimitationsSelected original resource https://arxiv.org/html/2412.04984v2; sha256:5bd418d4b53703bd12a2df4e8cdd51450675b111c9ec115d7e0b874bc1039db4; archived source capture; exact capture time not independently established
• Frontier Models are Capable of In-context Scheming · §5 ConclusionSelected original resource https://arxiv.org/html/2412.04984v2; sha256:5bd418d4b53703bd12a2df4e8cdd51450675b111c9ec115d7e0b874bc1039db4; archived source capture; exact capture time not independently established