Washington just blink-and-missed its way through a massive tech policy whiplash. Less than three weeks after slapping sudden export controls on Anthropic's crown jewels, Claude Fable 5 and Claude Mythos 5, the US Department of Commerce completely reversed course.
If you tried to log into Anthropic's most advanced systems over the last fortnight and found a digital brick wall, you weren't alone. On June 12, the Trump administration dropped a regulatory hammer, forcing Anthropic to suspend access to these frontier systems globally because it couldn't verify the nationality of its users in real-time. Now, Commerce Secretary Howard Lutnick says the restrictions are gone, and Anthropic is actively restoring global access. Meanwhile, you can read similar events here: Why Europe Is Moving Toward A Phased Social Media Ban For Kids.
But don't buy the clean corporate narrative that this was just a routine security check. This brief, aggressive ban reveals a messy, ongoing tug-of-war between federal national security hawks and the breakneck commercial realities of Silicon Valley.
The Secret Jailbreak That Panicked Washington
To understand why the government panicked, look at how these two models actually function. Fable 5 and Mythos 5 are built on identical core architecture, but they are tuned for completely different environments: To explore the bigger picture, check out the excellent report by Wired.
- Claude Fable 5: Built for general public and enterprise use, loaded with heavy safety layers to prevent bad actors from doing things like writing malicious code.
- Claude Mythos 5: Distributed through Anthropic's "Project Glasswing" to a tight circle of defense and corporate partners, stripped of several public guardrails specifically so cybersecurity teams can use it to hunt down software vulnerabilities.
The trouble started right after their June 9 launch. A trusted partner—reported to be Amazon—stumbled upon a narrow "jailbreak" technique. This workaround allowed users to trick Fable 5 into bypassing its safety filters, potentially exposing advanced coding capabilities that defense officials worried could be weaponized by state-sponsored hacking groups in China or Russia.
Instead of allowing Anthropic to patch the bug on the fly, the Commerce Department acted immediately, applying sweeping export controls. Because Anthropic had no structural mechanism to instantly screen out foreign nationals, it had to yank both models offline entirely, cutting off hundreds of millions of users and even its own overseas employees.
How Anthropic Bought Its Way Out of Regulatory Limbo
Anthropic didn't take the ban lying down. The company openly criticized the administration's heavy-handedness, arguing that if every minor jailbreak discovery resulted in a forced commercial recall, the entire American AI sector would grind to a halt. OpenAI CEO Sam Altman echoed the frustration, publicly noting that while safety checks make sense, he didn't like the idea of the government manually picking and choosing corporate customers.
The logjam finally broke when Anthropic built a specialized security classifier that explicitly targets and blocks the exact jailbreak method flagged by the government. Company sources indicate this new automated guardrail stops the exploit in over 99% of cases.
In exchange for lifting the export controls, Lutnick forced Anthropic to sign up for a much deeper, more intrusive relationship with Washington. The company has officially committed to:
- Proactively tracking and mitigating borderline security risks before they go public.
- Building shared pre-release testing protocols directly with federal agencies.
- Establishing an immediate reporting pipeline to alert the government to any detected malicious use of its systems.
The Bigger Geopolitical Stakes
This isn't an isolated bureaucratic spat. It happens against the backdrop of a broader executive order introducing a voluntary 30-day review window for advanced commercial AI models. White House Chief of Staff Susie Wiles explicitly thanked tech firms that cooperated with these new guardrail testing demands, signaling that the administration expects a level of subservience from top-tier labs.
It's also a delicate dance for Anthropic, which previously clashed with the Pentagon over attempts to prevent its technology from being used in autonomous weaponry. That dispute previously led to a temporary, controversial federal ban on Claude products that was ultimately blocked by a federal judge.
For now, the Commerce Department is keeping its hand on the trigger. Lutnick explicitly warned that the government retains the right to reimpose strict licensing requirements if Anthropic fails to live up to its security commitments or if global security conditions shift.
If you are a developer, researcher, or enterprise client relying on frontier systems, your short-term play is clear. Expect more friction, not less. While Fable 5 and Mythos 5 are back online today, the blueprint for federal intervention has been drawn. If your workflows depend entirely on a single proprietary model, you're exposing yourself to massive operational risk. Diversify your API integrations across multiple providers immediately, because the next sudden regulatory shutdown might not resolve itself in two weeks.
