Gallery

Contacts

405 W. Greenlawn Ave Lansing, Michigan 48910

contact@techjacksolutions.com

+1-616-320-4064

Skip to content
Technology Daily Brief Vendor Claim

AI Models News: Epoch AI Evaluates Fable 5 and Mythos 5, Independent Data Now the Only Public Record

3 min read Reuters Partial Moderate
Anthropic suspended global access to Claude Fable 5 and Mythos 5 on June 12, 2026, following a US government directive, and on the same day, Epoch AI published an independent evaluation of the models' capabilities. For developers and compliance teams, Epoch's findings are now the most complete public record of what these models could actually do.
FrontierMath Tier 4 (v2), 88% (Epoch AI, pending)

Key Takeaways

  • Anthropic suspended global access to Fable 5 and Mythos 5 on June 12, 2026, confirmed by Reuters, after a US government directive Anthropic couldn't comply with through targeted access controls alone.
  • Epoch AI published an independent evaluation of Fable 5 on the same day, reportedly finding 88% on FrontierMath Tier 4 (v2), pending URL confirmation, this is the only independent public record of the model's capability.
  • Anthropic disputes the jailbreak severity that reportedly triggered the government's action, stating the method doesn't bypass core safety classifiers, the jailbreak actor attribution remains unconfirmed.
  • Developers who had Fable 5 in active evaluation via Amazon Bedrock lost access before independent benchmarks cleared; substitution decisions should wait for Epoch's confirmed results.

Three days. That’s how long Fable 5 was publicly available before Anthropic pulled it offline.

Reuters confirmed on June 12 that Anthropic disabled global access to both Fable 5 and Mythos 5 after the US government ordered the company to suspend access. Anthropic cited an inability to filter users by nationality in real time as the reason it shut down access for everyone, not just foreign nationals. Per Anthropic’s own framing, a targeted restriction wasn’t technically feasible on the timeline the directive required.

The shutdown makes Epoch AI’s independent evaluation, published the same day, more consequential than it would otherwise be. Epoch assessed Fable 5’s performance on FrontierMath Tier 4 (v2), the benchmark suite designed to resist memorization through novel mathematical problems. According to the evaluation, Fable 5 reportedly achieved 88% on FrontierMath Tier 4 (v2), a figure that, if confirmed when the Epoch AI benchmark page resolves, would represent a meaningful independent signal rather than vendor-reported performance. That distinction matters. Anthropic’s system card (arXiv:2605.14153) also reports a 32% improvement on SWE-bench Verified and a 22% score on Agents’ Last Exam per UC Berkeley’s RDI evaluation, but those figures come through Anthropic’s own documentation, not independent reproduction. The SWE-bench figure is self-reported. The ALE figure references an independent evaluator cited within a vendor document.

The catch is this: the only benchmark with an independent evaluator, Epoch, is the one with a broken source URL. Until that page resolves, the 88% figure carries the same epistemic status as the vendor-reported scores.

The Wall Street Journal reported that Amazon CEO Andy Jassy contacted senior US administration officials, and those conversations triggered the escalation to a Commerce Department ban. Amazon’s relationship with Anthropic, a cumulative $13B investment per prior reporting, makes this stakeholder dynamic worth watching carefully. The specific legal authority behind the directive, and whether Secretary Howard Lutnick signed it directly, remains unconfirmed from available sources.

The jailbreak claim underlying the government’s action is itself disputed. Anthropic stated publicly that the demonstrated method “does not bypass the core safety classifiers that prevent harmful outputs,” per SecurityWeek’s reporting. The Wall Street Journal attributed the security concern to Amazon raising a flag through official channels. The specific actor, whether an Amazon cybersecurity team, an independent researcher, or both – hasn’t been confirmed from a single authoritative source. Don’t treat either characterization as settled.

What to watch

Two things resolve this story. First, the Epoch AI benchmark page: if it confirms the 88% FrontierMath result independently, it validates the capability claims in Anthropic’s system card and raises the stakes of the shutdown’s precedent. Second, re-access timeline: Anthropic hasn’t disclosed when, or whether, Fable 5 and Mythos 5 will come back online for international users.

TJS synthesis

If you were evaluating Fable 5 for production use via Amazon Bedrock, your evaluation window closed before the independent benchmarks cleared. Epoch’s June 12 brief is now the best public record of what the model could do. Don’t make migration or substitution decisions based on Anthropic’s system card alone, wait for the Epoch URL to resolve and confirm the FrontierMath result before benchmarking alternatives against it. The government’s action also sets a precedent that compliance teams should log: export control can now function as a real-time AI governance instrument, not just a procurement filter.

View Source
More Technology intelligence
View all Technology

Related Coverage

Stay ahead on Technology

Get verified AI intelligence delivered daily. No hype, no speculation, just what matters.

Explore the AI News Hub