METR’s mission to warn the public if AI gets close to going rogue

Chris Painter, president of Model Evaluation and Threat Research (METR), says the organization has focused since 2022 on ensuring the public would learn if AI were autonomous, difficult to steer and close to “going rogue.” He says METR has worked with OpenAI, Anthropic, Google DeepMind, Meta and Amazon on piloting third-party assessments and investigations. Painter also highlights findings that cut in different directions: research measuring rapidly improving AI capabilities, alongside a study showing software engineers using AI in early 2025 were slowed down even when they thought they were being sped up.