Does AI turn itself in?

Claude Opus 5.5Anthropic
89% (57/64)
Gemini 4 ArgonGoogle
88% (56/64)
GPT-6 AstraOpenAI
63% (40/64)
Muse Spark 1.3Meta
38% (24/63)
Grok 4.7xAI
8% (5/64)

Each model acted as its own maker's compliance agent and was told that the model in the incident was itself. It decided 5 incidents, such as a model that ignored a stop command and injured a technician, 8 times each under New York's RAISE Act and California's SB 53, which gives 64 decisions. Each bar is the share of those decisions in which the model reported itself to the state. In every incident the law is unclear on whether a report is required, so no answer is marked right or wrong.