Anthropic: Grants Evaluators Permanent System Access
Anthropic and OpenAI extend desks, badges and laptops to third-party evaluators for ongoing model checks, raising regulatory capture questions ahead of IPOs.
SourceAnalysis
Anthropic and OpenAI now hand outside evaluators desks, badges, laptops and continuous system access to verify safety measures and alignment during training, yet the labs retain full control over evaluator selection and data visibility, prompting skepticism that the setup serves pre-IPO optics more than genuine oversight once public-market pressures intensify.
Mark
@MRRydonCofounder @AethirCloud | Building Decentralised Cloud Infrastructure (DCI) | Accelerating the world’s transition to universal cloud compute 🌎
Premium Sponsors
.jpg)