Anthropic and OpenAI Split Over Massachusetts AI Audit Bill

Anthropic endorses strong independent audit provisions in a Massachusetts frontier AI bill while OpenAI is lobbying to weaken those same provisions, creating a public rift between the two companies. Rep. Casar called both companies' responses to his oversight letters 'insufficient.' Separately, Averi launched an eval methodology where the eval company never interacts with model weights and the model creator receives no information about the evaluation, enabling IP-safe third-party audits. Policy commentator Dean Ball argued that a false claim about OpenAI using 'neuralese' spreading widely on social media shows social media is structurally ill-suited to resolve technically complex AI safety questions, and called for legislated mechanisms where independent experts can scrutinize AI labs and report to governments and the public. Apollo Research successfully red-teamed Anthropic's auto-mode and plans similar campaigns with other AI companies.