Approximately 200 companies testing Anthropic's Mythos AI model through Project Glasswing have maintained their preview access despite a recent US government directive to shut it down, according to sources cited by Bloomberg. The firms, selected early by Anthropic to evaluate the model before a broader rollout, appear to have navigated the order's reach.
The situation pits corporate access against government authority, raising questions about how the US enforces restrictions on advanced AI systems. Project Glasswing, which grants early access to Mythos, became a test case for regulatory enforcement as authorities scrutinize emerging AI capabilities and their potential risks.
Bloomberg reports that the precise number of companies retaining access stands at about 200, though the shutdown order's scope remains unclear. Neither Anthropic nor US officials have commented on why the order was not fully enforced or whether further compliance actions are expected.
For these firms, continued access offers a competitive edge in developing AI-driven products, but it also exposes them to regulatory uncertainty. The discrepancy between the shutdown order and its implementation could signal challenges in governing rapidly evolving AI technologies.
The outcome suggests that even direct government orders may struggle to contain AI deployment when commercial interests and technical realities collide, underscoring the limits of top-down control in this domain.