Washington Forces Anthropic to Pull Its Top Claude Models

The News

The United States government has ordered Anthropic to switch off public access to its two most advanced models, Claude Mythos 5 and Claude Fable 5, in a move that stunned the company and rattled the wider AI industry. According to Anthropic, the directive arrived at 5:21 PM Eastern Time on Friday, 12 June 2026, framed as an export control measure citing national security.

The timing is striking. Fable 5, a guard-railed version of the flagship Mythos system, had gone live only three days earlier on 9 June, and had been ranked the most capable model available to the public on the Vals AI benchmarks. Mythos 5 itself, previewed in April, had never been released widely; access was limited under a programme called Project Glasswing to roughly 50 vetted organisations, including Amazon, Apple, Google, Microsoft and CrowdStrike.

At the heart of the dispute is the models' unusual talent for finding software security flaws. Anthropic says Mythos identified vulnerabilities in every major operating system and web browser it was tested against. The government's underlying worry, as the company understands it, is a claimed jailbreak that could coax Fable into reading a codebase and pointing out its weaknesses.

Anthropic is not hiding its anger. "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the firm wrote, describing the issue as a narrow, non-universal jailbreak.

Why It Matters

This is a watershed. Governments have fretted aloud about frontier AI for years, but ordering a private lab to pull a flagship product from the market is a different order of intervention. Anthropic, led by chief executive Dario Amodei, was widely expected to pursue an initial public offering in 2026, and a regulatory shutdown days after launch is exactly the kind of headline that unsettles bankers and investors.

There is bitter irony here too. Anthropic built its brand on caution, marketing Mythos as almost too dangerous to release, and that very positioning may have invited the scrutiny now disrupting its business. In April, OpenAI chief Sam Altman mocked the approach as fear-based marketing, likening it to announcing "We have built a bomb." Anthropic counters that the same vulnerability-hunting capability already exists in OpenAI's GPT-5.5, that cybersecurity teams use it defensively every day, and that applying this standard across the board would halt new model deployments entirely.

Indian Angle

For India, the episode lands at a sensitive moment. The export control framing means foreign nationals, including Indian developers and enterprises, are often the first to lose access when Washington tightens the taps. Thousands of Indian startups and IT services giants such as TCS, Infosys and Wipro build customer products on top of Claude through cloud platforms, and a sudden cut-off shows how fragile that dependency can be.

It also sharpens the case for sovereign capability. India's homegrown model builders, Sarvam and Ola's Krutrim among them, have long argued that critical AI infrastructure should not sit behind another country's national security veto. This is their strongest evidence yet, and it strengthens the pitch for the IndiaAI Mission's compute and model-building push.

Regulators should watch closely. MeitY is still shaping India's AI governance approach, and CERT-In has an obvious interest in models that can scan code for exploits. The affair is a live case study in how to treat dual-use AI: useful for defenders, dangerous in the wrong hands, and impossible to ignore.

FAQ

When did the shutdown take effect?

Anthropic says it received the government directive at 5:21 PM Eastern Time on Friday, 12 June 2026. The order required it to switch off public access to Claude Mythos 5 and Claude Fable 5 immediately, just three days after Fable 5 had been released to the public on 9 June.

Why were the models targeted?

The directive was framed as an export control action on national security grounds. Anthropic believes the real trigger is a claimed jailbreak that could let Fable 5 scan a codebase and surface security flaws. The company calls it a narrow, non-universal issue and disputes that it justifies a full recall.

What does this mean for Indian developers?

Indian startups and IT firms that build on Claude through cloud platforms face direct disruption when access is curtailed. The episode underlines the risk of depending on foreign frontier models and strengthens the argument for sovereign alternatives such as Sarvam and Krutrim, backed by the IndiaAI Mission.

This story was reported by TechCrunch. Read the full original coverage at TechCrunch.