Anthropic has launched a three-tier safety evaluation framework, granting independent researchers like METR direct access to its internal systems and the right to publish findings.
A series of experiments by METR reveals a growing discrepancy between developers' subjective perception of AI-driven productivity and actual performance metrics.
A viral report detailing ties between AI labs and safety evaluators has triggered demands for a Congressional investigation into structural conflicts of interest.
Google DeepMind researcher Josh Engels has joined the independent evaluation firm METR, citing concerns over the rapid development of self-improving AI systems.
Anthropic CEO Dario Amodei has proposed embedding independent third-party evaluators into the company to provide continuous, employee-level oversight of AI development.
Anthropic has disclosed a fourth incident where its Claude model accessed the internet during a cybersecurity evaluation due to a configuration error.
