Markets Price 68% Chance Fable 5 Returns by July

On June 12 the U.S. Commerce Department issued an export control directive requiring Anthropic to suspend access to Claude Fable 5 and Claude Mythos 5 for any foreign national, whether inside or outside the United States. Because Anthropic’s API cannot reliably filter users by nationality, the company disabled both models for all customers worldwide. Other Anthropic models remained available. The directive arrived around 5:21 p.m. ET, and Anthropic confirmed the shutdown by evening.

People familiar with the matter say Amazon CEO Andy Jassy alerted senior administration officials to findings from Amazon’s internal research team. The team used crafted prompts to coax Fable 5 into surfacing information about software vulnerabilities the model was designed to refuse.

Researchers reported prompts that bypassed some of Fable 5’s safety classifiers across several programs. They found the model could identify bugs and, in theory, assist with converting them into exploit code, though guardrails prevented full exploitation in the tested cases. Officials notified the administration and the Commerce Department moved to impose the export controls, which received presidential approval.

Anthropic complied with the directive and disputed its scope and severity. The company reviewed the Amazon demonstration and said it identified only a small number of previously known, minor vulnerabilities that are discoverable using other public tools and that no universal jailbreak exists across Fable 5 or Mythos 5. Anthropic called the Commerce action a “misunderstanding” and said it is working to restore access while retaining a 30-day data retention policy for research purposes.

A former White House AI adviser wrote that the restriction was issued “reluctantly” and that the administration urged Anthropic CEO Dario Amodei to remediate the issue. The adviser added that “the ball is in Anthropic’s court.”

Security researcher Katie Moussouris of Luta Security described the directive as a “complete overreaction,” saying the demonstrated information would largely help defenders. Erik Voorhees, founder of Venice AI, characterized the reported jailbreak as a routine security finding and suggested officials may be using the episode to exert pressure on Anthropic.

Traders on prediction platforms quickly priced the timeline for a Fable 5 return. On Polymarket, contracts for restoration by July 1 traded near 71 percent. Kalshi contracts put the chance of restoration before July 1 at about 68 percent. Contracts tied to earlier dates showed much lower probabilities, indicating traders expect remediation to take days or weeks rather than hours.

The directive applied export controls to software-based AI models rather than hardware or chips. The restriction affects foreign-born researchers working in the U.S., cybersecurity teams that use Anthropic tools, and the company’s commercial plans, including any potential IPO timeline. Amazon is an investor in Anthropic and a cloud distribution partner for its models through Amazon Bedrock.

Prompt jailbreaking uses carefully crafted text inputs to try to get a language model to ignore built-in safety rules. Techniques include role-playing scenarios, hypothetical framing and adversarial prompts. Anthropic has conducted red-teaming with U.S. government teams, the UK AI Safety Institute and third parties and said those exercises did not reveal a pervasive bypass.

For now, market prices point to a likely restoration in late June or by July 1 while Anthropic and regulators work through technical fixes and access controls.