The current AI policy moment is more concrete than the usual debate over ethics. Axios reported that OpenAI's GPT-5.6 received a federal green light for broad release after testing and meetings. AP reported that Anthropic's Fable 5 restrictions were lifted while Mythos 5 remains limited to approved U.S. organizations. Illinois, meanwhile, has enacted a law requiring frontier-model frameworks, incident reporting, and annual third-party audits.
Taken together, those developments show a shift from principle to access control. Governments are not merely asking companies to be responsible; they are deciding which models can be released, to whom, and after what kind of review. That may be necessary when models have cybersecurity capabilities that could be misused. It also creates a democratic problem if the evidence behind release decisions stays mostly private.
The strongest argument for case-by-case oversight is speed. AI systems evolve faster than legislation, and a rigid statute may age badly. The strongest argument against it is legitimacy. Private negotiations between a company and federal officials can shape markets, research access, and geopolitical advantage without the transparency normally expected of major public policy.
A durable framework should separate three questions: technical capability, misuse risk, and market fairness. A model may be powerful but manageable, risky for some users but safe for others, or restricted in ways that advantage one lab over another. Treating those as one question invites both overreaction and capture.