US lets Anthropic reopen Mythos 5 to select firms

The U.S. has let Anthropic reopen Mythos 5 to selected cybersecurity firms.

Anthropic’s latest move matters because it shows how tightly frontier AI access is now being managed: one model was blocked on June 12, then partially restored two weeks later, with more than 100 companies and institutions expected to get access.

1. Mythos 5 is back, but only for a narrow group

Anthropic said the U.S. government told it that Mythos 5, its strongest cybersecurity model, can be redeployed to a set of U.S. organizations that defend critical infrastructure. That means the model is not returning to everyone, only to a screened group the government considers trusted.

The practical takeaway is simple: access is now tiered. Anthropic is restoring use quickly for a limited set of firms while it keeps working with the government on broader approval.

• Model: Claude Mythos 5
• Use case: cybersecurity and critical infrastructure defense
• Access model: limited redeployment, not public release

2. The June 12 block is still the key turning point

This story starts with a rollback. On June 12, Anthropic abruptly pulled Mythos 5 and Fable 5 from all users after a government order restricted their release. The new decision does not erase that action, it only narrows the restriction.

That sequence matters because it shows how fast policy can change for frontier models. A model can move from general availability to a restricted pool in days, then back again for a subset of users once officials are satisfied.

• June 12: release restricted
• June 27: partial access restored
• Scope: U.S. organizations only, with vetting still unclear

3. The government is treating AI access like a security filter

The Trump administration said the concern was that powerful AI systems could be misused by military intelligence users in China, Russia, or elsewhere. That is why the rules focused on who may receive the model, not just on what the model can do.

The open question is how the vetting works. The government has not said which companies qualify or what criteria will decide access, and that uncertainty is now part of the policy itself.

• Stated concern: misuse by foreign military intelligence users
• Policy tool: controlled release, not blanket ban
• Unknown: selection criteria for approved firms

4. Anthropic wants Mythos 5 back in more hands

Anthropic said it is restoring access quickly and continuing to work with the government to expand access to Mythos 5. It also wants Fable 5 returned to general use, but that has not happened yet.

For Anthropic, the partial restart is both a business win and a signal that its relationship with regulators remains fragile. The company can serve some customers again, but it still depends on government approval for the rest of the rollout.

• Near-term goal: restore access for approved cybersecurity firms
• Next goal: expand access beyond the initial set
• Separate issue: Fable 5 remains outside general release

5. The decision puts pressure on rivals too

OpenAI chief Sam Altman echoed concerns about government selection of customers, saying extensive safety testing is reasonable but he dislikes the government picking customers. That comment shows the ruling is being watched well beyond Anthropic.

Reuters reported that more than 100 companies and institutions will now have access to Mythos 5, including many Fortune 500 companies, though the exact list is still unclear. In other words, the model is moving from a full stop to a controlled expansion.

• Industry effect: other frontier AI firms are watching the precedent
• Access pool: more than 100 companies and institutions
• Known users: many Fortune 500 firms, according to a source

How to decide

If you care about cybersecurity use cases, Mythos 5 is the item to watch because it is being restored first to defenders of critical infrastructure. If you care about policy, the more important story is the vetting process, since the government has not explained how firms will be chosen.

If you track AI markets, the bigger signal is that frontier model access can now be paused, screened, and reopened in stages. That makes Anthropic’s partial release a useful case study for how the next wave of AI distribution may work.