Anthropic gives 6 METR researchers access to inspect safety practices
Technology
Anthropic is letting six researchers from METR dig deep into its AI models and safety practices, giving them employee-level access.
This rare move comes after recent worries about unpredictable AI behavior, and it's all about making sure their tech stays safe and accountable.
METR can publish findings without approval
The METR team gets to use Anthropic's own tools and laptops to run their checks.
They can publish their findings without needing company approval first (unless customer data needs protecting).
Anthropic says this level of openness is inspired by how banks are regulated, showing they're serious about transparency, something CEO Dario Amodei has been pushing for in the whole AI industry.