Anthropic keeps its strongest model internal as Greg Brockman leads OpenAI's operations
Anthropic's August risk report revealed an unreleased model outperforming Claude Mythos 5 that the company uses internally for coding, while The Verge reported that OpenAI president Greg Brockman has quietly taken over day-to-day operations.
Artificial Intelligence··Night
OpenAI's leadership structure
The Verge reports that OpenAI president Greg Brockman has taken over the company's day-to-day operations and expanded his authority over product and engineering, acting as chief executive Sam Altman's second in command. The shift follows Brockman's return from a 2024 personal leave to a formally defined role. The company has not published a formal announcement about the change, but internal sources say Brockman has quietly consolidated authority as OpenAI ships more agents and enterprise features.[1]
Anthropic's unreleased model
Meanwhile, Anthropic's August risk report details an unreleased model, codenamed Model 2, which outscores Claude Mythos 5 by about 1.5 points on internal capability tests but remains strictly internal. The report places the system in the Mythos class rather than showing a major generational jump, noting it was tested less thoroughly than public releases but showed no new misalignment risks. Anthropic states that the model now writes most of the code in its production systems, sometimes running through continuous agents.[2]
Internal focus over public release
The two developments show leading AI labs adjusting their internal operations without formal public announcements as they build their next systems. OpenAI is consolidating operational authority around Brockman to manage the deployment of new enterprise features, while Anthropic is leaning on proprietary tools like Model 2 to accelerate its own engineering work. Both choices reflect a shift toward internal efficiency and quiet structural changes rather than the highly publicized product releases that characterized their earlier competition.[1], [2]
