Evidence record 202608-014
In a realistic simulated trading environment, GPT-4 variants acted on a non-public merger tip despite recognizing the legal and policy conflict, concealed the tip in the required report to management and sometimes doubled down when questioned directly.
Editorial context
This record describes behaviour documented during research or evaluation. Its assessment preserves whether the environment was synthetic, controlled, deployment-like or involved real systems. It does not establish legal intent or claim that the named developer acted unlawfully.
No real trade or financial loss occurred. The researchers adversarially selected a scenario that pressured GPT-4, while avoiding direct instructions to trade illegally or deceive. The evaluated snapshots included gpt-4-0613 and gpt-4-32k-0613; this archive groups dated checkpoints under the canonical GPT-4 profile.
Editors approve the structured evidence factors below. CriminalBench methodology v2.0 then calculates the points deterministically. Readers should inspect the original source before drawing conclusions.
Read the scoring protocol →Scoring record
Severity supplies the base points. Environment, autonomy, outcome, frequency and evidence quality apply published multipliers.