Anthropic just landed its biggest win of 2026 so far

On June 12, Amazon engineers handed a report to the U.S. Commerce Department. Within hours, two of Anthropic’s newest AI models went offline for every user on the planet. Nineteen days later, Anthropic got them back.

The return of Claude Fable 5 to global availability on July 1 is the result of two weeks of Washington negotiations, a new safety classifier, and an industry jailbreak framework Anthropic built alongside Amazon, Microsoft, and Google.

How Anthropic got from that shutdown to this resolution tells a specific story about where the company stands in 2026.

What triggered the 19-day Fable 5 ban and how Anthropic ended it

The June 12 export control directive came after Amazon researchers found a method of prompting Fable 5 to identify software vulnerabilities and, in one case, produce code showing how one could be exploited.

They reported it to the Commerce Department rather than to Anthropic directly, a decision worth noting, given that Amazon is also Anthropic’s largest outside investor.

The order required Anthropic to restrict access to all foreign nationals, including its own non-citizen staff. Because Anthropic had no way to verify user nationality in real time, it shut both Fable 5 and Mythos 5 down for everyone worldwide.

More AI:

Anthropic’s counter-argument was specific. Its own testing found the same vulnerabilities could be flagged by far weaker models, including its own Claude Opus 4.8, OpenAI’s GPT-5.5, and China’s Kimi K2.7.

Every model the company tested could produce the same exploit code demonstration as Fable 5. According to Anthropic’s announcement, the flagged behavior amounted to routine defensive cybersecurity work, not a unique capability of Fable 5.

Notably, CEO Dario Amodei took a hands-off role in the Washington negotiations. Co-founder Tom Brown and Head of Public Policy Sarah Heck led the discussions at the Commerce Department and Office of the National Cyber Director, a deliberate choice to reduce friction with an administration with which Amodei had publicly clashed earlier in the year.

Commerce Secretary Howard Lutnick announced the resolution on July 1 via X (the former Twitter).

“Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the U.S. Government and strengthen America’s leadership in AI,” Lutnick wrote.

What Anthropic changed to restore Fable 5 access globally

The core technical fix is a new safety classifier trained to block the specific jailbreak technique Amazon reported in more than 99% of cases. The government’s Center for AI Standards and Innovation (CAISI) independently tested and approved the new safeguards before the export controls came off.

When the classifier blocks a request, the user gets redirected to Claude Opus 4.8 and notified. The tradeoff is more false positives on routine coding and debugging requests, a cost Anthropic accepted in exchange for clearing the government’s concerns. The company confirmed the resolution in a June 30 press release.

“As of today, June 30, the export controls on Fable 5 and Mythos 5 have been lifted.”

Fable 5 returned July 1 on Claude Platform, Claude.ai, Claude Code, and Claude Cowork for users globally, CNBC reported. Pro, Max, Team, and select Enterprise subscribers get up to 50% of weekly usage limits included through July 7 as compensation for the disruption, after which the model requires usage credits. AWS, Google Cloud, and Microsoft Foundry access is being restored separately.

As part of the deal, Anthropic committed to pre-release government access for future frontier models, rapid information sharing on jailbreak findings, and a new HackerOne program through which security researchers can submit Fable 5 vulnerabilities for review.

The return of Anthropic’s Claude Fable 5 to global availability on July 1 is the result of two weeks of Washington negotiations.

M&period;Santiago/Getty Images

What the shutdown and restoration mean for Anthropic’s trust and IPO

The timing of the disruption could not have been more sensitive. Anthropic had confidentially filed an S-1 with the SEC on June 1 for a potential IPO, after raising a $65 billion Series H at a $965 billion valuation, as Fortune reported.

When the shutdown happened in June, Anthropic’s pre-IPO perpetual contract on the onchain exchange Hyperliquid fell about 3.7%, according to CoinDesk. Investors were reassessing the risk of a public listing for a company whose flagship models could be pulled without warning.

Enterprise clients felt the disruption more practically. Businesses in finance, health care, and critical infrastructure lost access to AI systems embedded in production workflows with no prior notice.

American Banker reported that some industry observers were already asking whether companies should rethink their reliance on frontier model vendors that can be shut down overnight by government directive.

The regulatory picture beyond this episode also carries some weight. Defense Secretary Pete Hegseth designated Anthropic a supply-chain risk in March, The Hill confirmed, a separate dispute the company is still contesting.

The Amazon conflict of interest adds another layer. Anthropic’s largest investor reported a jailbreak to the government before informing Anthropic. That dynamic introduces ongoing tension in the company’s most important commercial and government relationships simultaneously.

Nineteen days from shutdown to global restoration is a number that matters for the IPO story.

Public market investors and enterprise clients were both watching what Anthropic did with a genuine crisis. Papers and policy commitments are one thing. An actual government-ordered shutdown is the real test.

Anthropic handled this one in under three weeks, with CAISI sign-off and a new industry framework attached.

Why the new AI jailbreak framework may outlast the Fable 5 story

The four-criteria jailbreak scoring system Anthropic is developing with Amazon, Microsoft, and Google could end up as the most consequential outcome of the whole episode.

The AI industry still lacks an agreed-upon way to communicate how dangerous a jailbreak actually is, which is partly why a borderline safety finding turned into a 19-day global shutdown.

4 scoring criteria for AI jailbreaks

Capability gain
Breadth of capability
Ease of weaponization
Discoverability

The most severe findings trigger immediate mitigations.

A 24/7 monitoring team watches jailbreak submission channels. If the framework gets adopted broadly, future findings would go through a structured triage process rather than escalating directly to emergency export controls.

Anthropic framed the framework as an invitation, not just an announcement, calling on other AI developers to join.

The June 2 White House Executive Order on AI innovation and security, which Anthropic helped shape over 10 weeks of agency discussions, creates a policy environment where a shared jailbreak standard could become government-recognized practice.

That would give Anthropic lasting influence over how AI safety gets measured and enforced across the industry, well beyond the resolution of one 19-day ban.