Anthropic’s most powerful consumer software tool went dark on June 12 after the Trump administration imposed surprise export controls with just 90 minutes’ notice. Eighteen days of tense negotiations later, the Commerce Department has backed down, and users worldwide regained access on July 1. The episode may signal the beginning of an era where governments decide who can use powerful software, and when.
On the evening of June 30, Anthropic posted four short sentences on X. The Department of Commerce had lifted export controls on Claude Fable 5 and Mythos 5. Access would be restored the next day. The company thanked its users for their patience.
Those four sentences ended one of the strangest standoffs in recent technology history, one that left millions of paying subscribers locked out of a product they had already purchased, rattled an industry unsure of where the government’s authority ended, and handed Chinese competitors a useful window of time to close the gap.
Anthropic launched Claude Fable 5 and Mythos 5 in early June. Fable 5 was the company’s first genuinely advanced model released to the public. Mythos 5 remained restricted to a smaller group of vetted organisations. Both were immediately praised across industry benchmarks as among the most capable software tools available anywhere.
The good press did not last long.
According to the Wall Street Journal, researchers at Amazon discovered a sequence of prompts that could push Fable 5 into providing information useful for cyberattacks, specifically a technique known as a jailbreak. Amazon’s chief executive Andy Jassy apparently raised the concern with White House officials. On June 12, the Commerce Department issued a private letter to Anthropic imposing export controls on both models, barring any foreign national, regardless of location, from accessing either.
Anthropic had 90 minutes to respond. With no practical way to verify the nationality of every user on its platform in that window, the company took both models offline entirely. Millions of subscribers worldwide, including paying customers and researchers in allied countries, lost access without warning.
What followed was a fortnight of near-silence from Anthropic and growing frustration across the industry. Dario Amodei, the company’s chief executive, stepped back from leading the negotiations. Tom Brown, an Anthropic co-founder, took his place as the company’s main point of contact with the administration. Lutnick’s subsequent letters were addressed to Brown, not Amodei. The company later confirmed that Amodei’s well-known public criticism of the administration and his backing of Kamala Harris in the 2024 presidential election had complicated the relationship.
By June 26, partial progress had been made. The government cleared Mythos 5 for use by more than 100 approved American organisations, particularly those operating critical infrastructure. Fable 5 remained offline.
Four days later, the full resolution came. Commerce Secretary Howard Lutnick wrote on X that his department had spent two weeks working closely with Anthropic to analyse Fable 5 and ensure it met government requirements. “Over the past two weeks, we have worked closely with Anthropic to analyse and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI,” he wrote.
See Also: Mastercard cyber hub launches in South Africa and Nigeria
The company did not walk away empty-handed from the negotiations, but it did walk away with conditions attached. According to Anthropic’s own statement and reporting by Axios, the company agreed to three main commitments. It will proactively detect and address security risks associated with its models. It will work with the government on protocols for future releases. It will report any malicious activity it identifies in how its models are being used.
On the technical side, Anthropic says it built a new safeguard specifically targeting the jailbreak that Amazon’s researchers found.
The fix blocks the problematic sequence 99 per cent of the time. In the remaining cases, Fable’s output is limited to security vulnerabilities that are already publicly known or have already been patched, meaning no new harmful information is produced. The fix does carry a trade-off. Anthropic acknowledged the new safeguard may occasionally flag legitimate requests as suspicious, creating some friction for users with entirely innocent purposes.
The episode exposed something that the technology industry had not fully confronted before. A government can, apparently, order a company to take its software offline globally with less than two hours’ notice, citing national security concerns it declines to specify in detail.
Tanishq Abraham, a former research director at Stability AI, told Al Jazeera the resolution raised questions the industry has not yet answered. “The biggest question now is: What precedent does this set for the industry? Does the US government need to approve every frontier model release?” he said. The rollout of OpenAI’s GPT-5.6 last week, which was also limited to a small group of government-approved customers at launch, suggests the answer may be moving toward yes, at least for the most capable tools.
The episode also drew criticism from investors and executives who pointed out that blocking American software handed Chinese developers an uninterrupted window to close the gap. While Anthropic’s models were offline, Chinese open-source models continued to circulate freely. Zhipu AI’s latest release, which reportedly matched some of Fable’s security research capabilities, attracted growing attention during exactly the period when Fable was unavailable.
Fable 5 returned to the Claude platform, Claude.ai and Claude Code on July 1. It is also being restored on Amazon Web Services, Google Cloud and Microsoft Foundry as quickly as possible. For users on Pro, Max, Team and selected enterprise plans, Fable 5 will count toward up to 50 per cent of their weekly usage limits through July 7.
The Trump administration faces an August deadline, set by its own executive order signed earlier in June, to establish standardised benchmarks for evaluating the security risks of new software models before release. How that framework takes shape will likely determine whether the kind of improvised shutdown Anthropic just experienced becomes a repeatable event or an aberration.
For now, the models are back. The relationship between Anthropic and the government is described, by those close to it, as improving. Whether the next powerful tool that leaves a company’s servers triggers another 90-minute countdown is the question the whole industry will be watching.
About the Author
Marcel Chidozie is a tech analyst and writer covering foreign news, fintech, and emerging technologies at TechRegard. Based in Nigeria, He's passionate about translating complex tech developments into compelling, accessible stories for diverse audiences. His work focuses on how technology shapes innovation across Africa and globally.
