OpenAI GPT-5.6, Anthropic models: U.S. government control

GPT 5.6 will be released only into a limited preview, with the U.S. government approving access "customer by customer" until a general release can be approved. The move comes two weeks after the government pulled Anthropic’s Fable and Mythos models, and it places OpenAI and Anthropic in the same regulatory bind.

How will the GPT 5.6 release process work?

The initial rollout will be a limited preview and the government will greenlight access on a per-customer basis, a process described in the report cited by the article. OpenAI CEO Sam Altman reportedly projected that the preview could last a "couple of weeks," though the piece notes Mythos has already been in preview for months and shows no sign of an imminent general release.

The article highlights a practical bottleneck: a government approval process that vets frontier models customer by customer. That approach may require bespoke testing for each model and user, and there is skepticism within the piece about whether regulators have the expertise or capacity to run those evaluations.

What problems does this create for AI companies and the market?

The immediate commercial risk is clear: extended previews can significantly limit the economic upside of expensive new systems. The article argues that if a preview lasts longer than a "couple of weeks," as Altman suggested, it could blunt revenues from a costly model and slow the pace of data center expansion tied to new model rollouts. Mythos is already an example: it has been in preview for months, and there is no indication it will reach general release soon.

The story frames the issue as industry-wide rather than a dispute between two firms. OpenAI and Anthropic now face "the same exact position" with identical regulatory friction and the same potential downside if they fail to secure approvals. The piece warns that a haphazard approval regime would be costly to implement and would not advantage a single lab.

What are the regulatory and safety gaps?

The article notes real safety concerns that justify scrutiny: AI tools are changing cybersecurity, and analogous issues appear in biorisk and alignment. Still, it raises a central question about the mechanics of oversight. GMU fellow (and soon-to-be OpenAI employee) Dean Ball argues there is a lack of clarity on what safety assurances would satisfy regulators, and the U.S. government lacks the capacity and expertise to perform the kind of testing this would require. The piece adds that it is "not even clear what regulators would be trying to protect against," which complicates design of any release regime.

The author urges collective approaches: trusting independent groups to guide the process, accepting imperfect regulatory options, and coordinating industry responses rather than using regulation as a competitive lever.

Why it matters

This is not a narrow dispute between two labs. The same approval process will apply to multiple frontier models, so the cost and delay hit the whole industry. If model rollouts are bottlenecked by customer-by-customer approvals or prolonged previews, the economic case for large model investments and associated data center buildouts weakens. That could slow development and labor and capital flows tied to large-model deployment.

What to watch

Watch whether GPT 5.6 exits its limited preview in the "couple of weeks" Altman reportedly projected, and whether Anthropic’s Mythos moves out of its months-long preview. Also monitor any public articulation from regulators about the specific risks they intend to address and whether independent testing bodies are enlisted to assist with model evaluations.