Anthropic Grants Staff-Level Access for Frontier AI Model Evaluations

The notable shift here is structural rather than technical: frontier labs are ceding some evaluation visibility to outside parties, and the involvement of a consultancy like Accenture alongside a safety-focused evaluator suggests risk testing is being folded into enterprise-facing assurance. That matters for enterprises weighing deployment, since third-party assessments can become procurement criteria. Whether published results carry enough methodological detail to be comparable across labs, and whether other developers follow with similar access, is the open question.