Anthropic did not publish a "risk discussion" of Mythos when required by their RSP

Anthropic, the AI safety-focused company behind Claude, may have breached its own safety protocols with the launch of its new model, Claude Mythos. According to an analysis, the firm's operative Responsible Scaling Policy (RSP 3.0) required the publication of a discussion on how a new model changes its risk analysis. This requirement is triggered either within 30 days of a significant internal deployment or immediately upon a public release. The core issue is that Anthropic provided early access to Claude Mythos to its 'launch partners' weeks before the official announcement, an action that likely qualifies as a public deployment under the policy's intent, yet no corresponding risk discussion was published.

While the company might argue the 30-day internal deployment rule applies, the case for the immediate public release requirement appears stronger. The RSP does not explicitly define 'public release,' but releasing a powerful model to a select group of external partners with potential internet-wide access is difficult to classify as anything else. This incident points to ambiguities within the RSP framework itself that need fixing. More critically, it reveals a potential failure in Anthropic's organizational ability to follow its own safety checklists—a concerning lapse for a company whose brand is built on rigorous safety practices.