Anthropic plans a new policy framework on AI development pace, extending a safety debate beyond dangerous-model releases to the speed of research and its relation to safeguards for individual models.
In an Oct. 19 roadmap, the company said policies aimed at stopping dangerous model deployment may need an added element addressing how quickly AI development proceeds. The roadmap treats research speed as a separate policy question alongside safeguards for specific models.
Under Anthropic’s earlier Responsible Scaling Policy, training or deploying a model with catastrophic-harm risks depended on safeguards bringing those risks down to a level the company considered acceptable.
Anthropic’s February revision set out two tracks: steps for the company itself and broader proposals for the rest of the industry. The company said it would not commit to taking on those wider measures alone.
Anthropic also set Dec. 15, 2026, as its target to publish a policy proposal on government support for embedded third-party evaluators, including their funding, access and independence.
In September, Amodei argued for slower frontier development and shared safeguards across companies and governments, as EconoTimes reported.
Anthropic set Jan. 15, 2027, as the target date for releasing its pace framework, with an explanation to follow if it chooses not to publish.
