New Fingerprint Test Detects If an API Is Really Running the Model It Claims

A new open-source tool uses behavioral fingerprinting to verify whether an API endpoint is actually serving the model it claims. By analyzing distributions of one-token answers across repeated queries, the test achieves an equal error rate of about 10.6% with 8 test cells. The tool outputs four verdicts—match, uncertain, mismatch, or insufficient—and is available as an MIT-licensed TypeScript CLI.