Anthropic’s most advanced Claude models are coming back online after Washington briefly froze access over cybersecurity concerns.
Export controls on Claude Fable 5 and Mythos 5 have been lifted, allowing the company to restore access after a two-week freeze. Users are regaining access, while AI developers are getting a warning about how quickly the Trump administration can intervene when frontier models raise cyber-risk concerns.
Access may be returning, but the fight over high-powered AI releases is far from settled.
Washington reverses course
Anthropic said it would begin restoring access after receiving notice from the Department of Commerce.
“We’ll begin restoring access tomorrow, and will share an update soon,” the company announced, adding that it was grateful to users for their patience.
Under the earlier order, Anthropic had to restrict access for foreign nationals, including foreign-national employees. The company said it suspended access for all users because the directive took effect immediately and it had “no reliable way to verify nationality in real-time.”
Commerce Secretary Howard Lutnick said officials worked with the company to review Fable 5 and approve its release in line with US government concerns.
Amazon jailbreak finding drove the scrutiny
Fable 5’s safeguards became the focus of the review.
Fable 5 and Mythos 5 share the same underlying model, but Fable 5 carries stronger safeguards for broader use, while Mythos 5 has fewer safeguards and was limited to selected Glasswing partners for defensive cybersecurity work.
CNN reported that Amazon, identified as a trusted partner, discovered a jailbreak that could circumvent Fable’s guardrails. Anthropic pushed back on the severity, saying the jailbreaks were “simple” and similar to workarounds found in other publicly available models. The AI company later added a safeguard to block the vulnerabilities Amazon found.
The improved safety classifier blocks the Amazon-linked bypass in more than 99% of cases, Anthropic said. Stricter filtering may also flag benign coding and debugging requests more often.
Fable 5 returns to Claude and cloud platforms
CNBC said Fable 5 is expected to return to global users on Claude.ai and Claude Code, counting for up to 50% of weekly usage limits through July 7 for Pro, Max, Team, and selected enterprise plans.
Cloud access is also expected to resume, with Anthropic planning to re-enable Fable 5 on Amazon Web Services, Google Cloud, and Microsoft Foundry as soon as possible. Mythos 5 appears to remain more tightly controlled, with access restored for some US organizations and broader availability linked to Glasswing.
Security teams face the real trade-off
The same cyber concerns that drew Washington’s attention are also what make Fable 5 and Mythos 5 relevant to enterprise security teams. Fable 5 raises the question of how safely advanced AI can be offered to broader users, while Mythos 5 shows why more capable systems may stay closer to controlled cybersecurity programs.
CISOs now have a planning problem. If a high-capability model can be restricted over cyber concerns, companies cannot treat access as guaranteed once a tool is added to their workflow.
Security teams using frontier AI may need backup tools and clearer rules for which teams or regions can use the most capable models. The lesson from Anthropic’s freeze is that AI access planning now belongs inside AI risk planning.
The Trump administration’s request puts GPT-5.6 on a more controlled release path as government scrutiny of frontier AI grows.