Until June, the government's role in frontier AI was mostly hearings and frameworks. In June it directly controlled who could use two frontier models.
Fable 5 launches, then disappears
Anthropic released Claude Fable 5 on June 9. It is a Mythos-class model made generally available with safeguards, released alongside Claude Mythos 5, a restricted version with some of those safeguards lifted for vetted users.
Three days later, on June 12, the Commerce Department issued an export control order citing a claimed jailbreak with national security implications. The order barred access for any foreign national, inside or outside the US, including Anthropic's own non-citizen employees. Anthropic had no reliable way to check every user's nationality in real time, so it shut both models off for everyone.
The controls were lifted on June 30, and Fable 5 came back on July 1. It was offline for 18 days.
OpenAI gets the same treatment, more quietly
On June 26, OpenAI limited its new GPT-5.6 family (Sol, Terra, and Luna) to a small group of trusted partners. The request came from the White House Office of the National Cyber Director and the Office of Science and Technology Policy, which were building a framework for security testing new models. OpenAI said it did not think government access reviews should become the long-term default. The public release came on July 9.
Model availability is now a vendor risk
This is the lesson from June for every company building on these models. A team that standardized on Fable 5 on June 9 lost it on June 12, with no warning and no timeline. The cause was policy, and it can happen again.
The answers are ones infrastructure teams already know:
- Keep a fallback model tested and ready. Switching providers should be a config change.
- Run your evals against the fallback continuously, so you know what quality you give up when you switch.
- Record which model served every request. When a model disappears mid-quarter, you need to know which workflows depended on it. That is an observability problem, and most teams can't answer it today.
Everyone files for an IPO
Anthropic confidentially submitted a draft S-1 on June 1. OpenAI announced on June 8 that it had done the same, saying it expected the news to leak anyway. Also on June 1, Alphabet announced an $80 billion equity raise for AI infrastructure, later upsized to $84.75 billion including a $10 billion private placement with Berkshire Hathaway.
Alphabet raising equity is the more surprising move. It generates enormous cash from ads and still decided it needed outside capital to keep up with compute demand.
What June tells us
Frontier releases now have a government gate, at least for models with serious cyber capability. Treat that like any other supplier risk. Assume a model can vanish, and make switching cheap before you need to.