Anthropic’s Mythos 5 Access Returns for Trusted Cyber Defenders

The US government has cleared Anthropic to restore access to its powerful Mythos 5 AI model, but only for a small group of trusted cyber defenders. This comes after months of tight restrictions due to national security concerns.

Mythos 5, Anthropic’s most advanced cybersecurity AI, was initially launched in April as part of Project Glasswing. About 200 firms across more than 15 countries received early access. The model helped identify over 400 Firefox bugs for Mozilla in just one month. But in June, the US Department of Commerce issued an export-control directive that forced Anthropic to disable Mythos 5 and its other frontier model, Fable 5, worldwide.

Anthropic shut down all access on June 13, 2026, because it could not reliably tell if users were domestic or foreign in real time. Commerce Secretary Howard Lutnick later wrote to Anthropic, saying Mythos 5 could be released to “certain trusted partners” after negotiations. The letter made no mention of easing restrictions on Fable 5, which remains offline.

The National Security Agency (NSA) lost access after the June 12 directive. However, the agency had already tested Mythos 5 in a controlled red-team exercise. Sen. Mark Warner described the model as breaking into “almost all of our classified systems, not in weeks, but in hours.” U.S. officials clarified that Mythos identified vulnerabilities quickly but didn’t necessarily exploit them right away.

Why the Restrictions Matter

Anthropic’s Mythos 5 is built to find cybersecurity weaknesses fast. It’s powerful enough to change how cyber defense and offense work. The Five Eyes intelligence alliance warned that frontier AI models like Mythos could transform cyber operations within months.

But the US government sees risks. It ordered Anthropic to limit Mythos 5 and Fable 5 access to US citizens only. That blocked global users, disrupting cybersecurity research worldwide. Anthropic also intentionally made Mythos 5 less helpful when it detects AI research. This was to slow down rapid AI development without safety controls.

European regulators have pushed back. They called for technological sovereignty and worry that US restrictions could hinder AI progress. Gina Neff, a professor of responsible AI at Queen Mary University London, warned that limiting access could stifle development and collaboration. She noted UK tests showed Mythos could bypass defenses 73 percent of the time.

Anthropic’s Ongoing Challenges

Anthropic has been at odds with the US government since early 2025. The Department of Defense labeled it a “supply chain risk,” and Anthropic is suing the Pentagon over that designation. Meanwhile, Anthropic reviewed a jailbreak demonstration and found only minor vulnerabilities in Mythos 5.

Since the shutdown, Anthropic issued refunds to affected users, including clients like Apple, with a June 20 deadline. By June 22, users regained access to Fable 5 through Claude Code apps, but Mythos 5 stayed offline for most. The partial restoration announced on June 27 allows only a small group of trusted cyber defenders to use Mythos 5.

Anthropic has also filed confidentially for an IPO valued above $900 billion. This shows the company’s high-tech ambitions despite regulatory setbacks. The path forward will likely require balancing AI innovation with national security concerns.

The Mythos 5 saga shows how powerful AI is reshaping cybersecurity. Governments want to control access to these frontier tools. At the same time, researchers and companies want to push AI limits safely. For now, Mythos 5 is back for a select few. But the debate over who gets to use these models is far from over.