The Commerce Department lifted its June 12 export controls on Tuesday night after Anthropic agreed to detect security risks, coordinate future releases with the government, and report malicious activity. Fable 5 began returning to users worldwide on July 1, nineteen days after both models went dark. Anthropic also shipped a new safety filter it says blocks the disputed jailbreak more than 99 percent of the time.

One of the strangest standoffs in the short history of AI regulation ended with a two-sentence post on X.

"We'll begin restoring access tomorrow, and will share an update soon. We're grateful to our users for their patience, and to everyone who worked with us on redeploying the models." — Anthropic (X, June 30, 2026)

Behind those two sentences sit nineteen days of dead API endpoints, a Senate account of an AI that reportedly walked through classified systems in hours, accusations of political retaliation, and a pressure campaign from an unexpected direction: Asian labs shipping rival models while America's most capable AI sat in a locked room.

On Tuesday night, the US Department of Commerce notified Anthropic that it had removed the export controls on Claude Fable 5 and Mythos 5. Both the company and the department confirmed the decision within hours. Fable 5, the public model, began coming back online worldwide on Wednesday, July 1. Mythos 5, the restricted version, returns to the vetted organizations that had it before.

The blackout is over. The questions it raised are not.

Washington Blinked on a Tuesday Night

The original directive landed on June 12, when the government added both models to its list of export-restricted technologies. That meant no access for any foreign national without special approval, a condition Anthropic could not enforce at scale. So the company pulled both models for everyone on the planet, a decision covered in detail in LDS's earlier reporting on the shutdown.

Now the license requirement is gone. Commerce Secretary Howard Lutnick framed the reversal as the outcome of a negotiation rather than a retreat.

"Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI." — Howard Lutnick, US Commerce Secretary (X, June 30, 2026)

What Anthropic gave up to get its models back fits in a single sentence. According to Lutnick, the company "has agreed to proactively detect and address security risks associated with the models; to work diligently with the U.S. government on protocols and standards and releases for Mythos, Fable and future models; and to inform the US government of any malicious activity."

There is a wrinkle in that list. As TechCrunch noted, Anthropic had publicly pledged to do much of this voluntarily, months before the export rule existed.

The restoration itself is broad. Fable 5 is returning to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. On Pro, Max, Team, and select Enterprise plans, Anthropic is including Fable 5 for up to 50 percent of weekly usage limits through July 7, after which it shifts to usage credits.

Anthropic Shipped a Fix for a Flaw It Still Disputes

The government's stated concern traced back to a report by Amazon researchers who claimed they had found a way to bypass Fable 5's safeguards and get the model to identify software flaws that could be exploited. Anthropic disputed the report's significance from the start, arguing the bypass surfaced only known, minor issues that other public models could find without any trick at all.

The company still built a remedy. In a blog post published Tuesday night, Anthropic said it trained a new safety filter that blocks the reported workaround more than 99 percent of the time.

It went further. Anthropic said it is now working with Amazon, Microsoft, Google, and its other Project Glasswing partners on a common framework for scoring AI jailbreaks, including thresholds for when a bypass is serious enough to demand new safeguards. Glasswing is the limited-access program through which Anthropic first released Mythos to vetted security teams at Apple, Microsoft, and Amazon earlier this year.

The subtext is hard to miss. The next time someone claims a jailbreak, there will be an industry rubric for deciding whether it matters. The last one cost Anthropic nineteen days of global availability.

How Nineteen Days Unfolded

JUNE 12, 2026

The Friday letter arrives

Commerce orders Anthropic to cut off foreign access to Fable 5 and Mythos 5, citing national security authorities. Unable to comply selectively, Anthropic pulls both models worldwide within hours.

JUNE 15, 2026

Security researchers turn on the order

TechCrunch reports that cybersecurity experts believe the ban "was never about an AI jailbreak" and looks more like political pressure than a security fix.

JUNE 21, 2026

A Senate account recasts the story

Senator Mark Warner relays an NSA account that Mythos broke into almost all of the agency's classified systems in hours during an authorized red-team exercise.

JUNE 25, 2026

China's open models close the gap

Axios reports that Chinese open-source models like GLM-5.2 are beginning to show cyber capabilities similar to those of major US labs.

JUNE 26, 2026

A partial thaw

The administration partially lifts the ban on Mythos 5, letting roughly 100 vetted organizations regain access. Fable 5 stays dark for everyone.

JUNE 27, 2026

Rivals fill the vacuum

TechCrunch reports Asian startups are launching "Mythos-like" models, among them Fugu and Tulonfeng, as the export ban drags on.

JUNE 30, 2026

The controls come off

Commerce notifies Anthropic the export controls are lifted. Lutnick and Anthropic confirm the deal on X that night. Anthropic publishes its redeployment plan and new safety filter.

JULY 1, 2026

Fable 5 comes back online

Access begins restoring worldwide on the Claude platform, Claude.ai, and Claude Code, nineteen days after the models went dark.

Rival Models Did What Lobbying Could Not

The timing of the reversal was not an accident of paperwork.

While Fable 5 sat offline, Asian AI companies began releasing models approaching Mythos-level capability, among them Fugu and Tulonfeng. Two days before the partial thaw, Axios reported that China's GLM-5.2, whose open weights already beat GPT-5.5 at coding, was showing cyber capabilities comparable to frontier US systems. Every week the ban continued, the case that it protected anyone got weaker.

The ban was also hurting the government itself. Nextgov reported that parts of the NSA, which had been using the latest Mythos build, lost access under the order the administration had written.

Danny Jenkins, CEO of the security firm ThreatLocker, put the practitioner frustration bluntly:

"Stronger AI models aren't a genie that can be crammed back into a bottle, no matter how much we might like to. All of this means that the only ones being restricted by the export controls are organizations that desperately need to test their systems and code." — Danny Jenkins, CEO of ThreatLocker (Nextgov, July 1, 2026)

The Review Regime Survives the Reversal

The models are back. The system that took them away is still standing.

A June executive order established a voluntary framework that gives the government early access to frontier models for up to 30 days before public release. Influential analysts criticized it, including Dean W. Ball, the former White House AI policy adviser who joins OpenAI this month to lead its new Strategic Futures team.

The pattern is already spreading. OpenAI said Friday it will initially limit access to three of its GPT-5.6 models to select partners after conversations with government officials, while stating it does not want government access reviews to become the long-term default. Whether OpenAI gets a say in that is an open question.

For working engineers, the practical lesson of June is uncomfortable: availability of a frontier model in your stack is now subject to a Friday letter from Commerce. Teams that treated claude-fable-5 as infrastructure spent nineteen days rediscovering the difference between a dependency and a guarantee.

The Other Side

The administration's defenders can point to a process that, on paper, worked. A researcher-reported bypass triggered a review, the government and the company negotiated safeguards, a stronger filter shipped, and an industry-wide jailbreak framework emerged. Lutnick's account describes coordination, not coercion.

The skeptics' reading is darker. Anthropic had already promised most of the lifted deal's terms voluntarily, the bypass surfaced flaws Anthropic characterized as previously known and minor, and the two-week resolution arrived only after Chinese rivals started filling the gap. On that reading, the export control was a lever in a broader feud between the administration and a company whose executives have publicly criticized it, one that has already produced a Pentagon blacklist and ongoing litigation.

Both readings fit the public facts. The government has not released the analysis behind either the ban or its reversal.

The Bottom Line

Strip away the noise and this is what happened: Washington pulled what are widely considered the most advanced AI models on the market offline globally with no court order and no public explanation, kept them dark for nineteen days, and handed them back in exchange for commitments their maker says it had already made.

Anthropic got its models back. The government got its framework. The industry got a precedent: frontier model availability is now, in practice, negotiable with the state, and every lab watching adjusted its release plans accordingly.

As Jenkins put it, stronger AI models aren't a genie that can be crammed back into a bottle. Washington spent nineteen days trying anyway, and nobody involved is promising it won't try again.