Anthropic's Glasswing precedent exposes AI access governance gap

The Glasswing precedent set by Anthropic reveals a critical governance gap in frontier AI. When they developed the Mythos model—which possessed powerful offensive capabilities—Anthropic unilaterally decided which organizations and state-adjacent institutions could access it. Without agreed criteria, external accountability, or democratic legitimacy, a single private lab became the arbiter of who could use strategically important technology and who could defend against it. The article argues this arrangement could have significant consequences, including increased coup risk, and that such decisions should be made by a governing body, not whichever lab happens to develop the most capable model.

The article develops the criterion of democratic resilience as an example of what should guide access decisions. However, even this reasonable criterion generates decisions of such political complexity that no private actor has the legitimacy to make them. Anthropic now functions as a geopolitical actor with structural power, triggering obligations to uphold international governance norms—a responsibility they did not choose. The author concludes that a regulatory framework is needed, where labs make decisions based on agreed rules and are accountable to an external body. Without this, the governance gap will only widen as future, more capable models emerge from less safety-conscious teams.