Anthropic’s New Model, The Mythos Wolf, Glasswing and Alignment
Anthropic has declined to release its new model, citing safety concerns about potential dangers, though commentators express skepticism about the reasoning. The move raises broader questions about responsibility in large language model development and the criteria governing public release decisions.
The story follows Anthropic’s announcement of a significant computing infrastructure deal—a 3.5GW capacity agreement involving partnerships with Google and Broadcom for TPU sourcing. This expansion suggests the company is preparing substantial computational resources for training and deployment, making the withholding of the new model more notable.
The model’s safety classification and refusal to release it publicly reflect ongoing industry tensions between accelerating AI capability development and managing alignment and safety risks. Anthropic’s position as a leading AI safety-focused company adds context to this decision, positioning safety considerations as paramount even when release might be commercially advantageous.