METR (Model Evaluation and Threat Research) has collaborated with OpenAI, Anthropic, Google DeepMind, Meta and Amazon since 2022 on third-party evaluations. President Chris Painter said METR does not accept funding from frontier AI companies or their employees, relying instead on diverse donors while using free model access supplied by companies. The nonprofit aims to disclose internal evidence to the public and government if AI becomes unmanageable. Its reports have identified both rapid capability gains and cases in which AI abilities were overstated. METR operates under non-disclosure agreements but seeks to keep public contract terms and conflict-of-interest disclosures. Painter said current third-party evaluations remain voluntary and offer limited oversight.