Anthropic rolls out public version of Mythos AI model with restrictions on cybersecurity use

Claude Fable 5 gives the public a taste of Anthropic's most powerful model, but the real cyber-offense capabilities stay behind locked doors.

Share

Add us on Google by Editorial Team Jun. 9, 2026

Anthropic just split its most advanced AI model into two distinct products, and the dividing line is cybersecurity. Claude Mythos 5, the full-power version capable of discovering and exploiting software vulnerabilities, remains restricted to vetted organizations. Claude Fable 5, the public and enterprise variant launched on June 9, 2026, ships with safety guardrails firmly in place and the cyber-offense toolkit stripped out.

Two models, very different access levels

Mythos 5 represents a major leap in AI-driven vulnerability research. The model builds on the Mythos Preview released in April 2026, with substantial gains in its ability to identify and exploit software flaws. During an initial testing phase called Project Glasswing, around 40 organizations were given limited access to evaluate the model’s capabilities. The results were apparently compelling enough that Anthropic expanded the program before settling on the restricted-access commercial launch.

Fable 5, meanwhile, is the version most people will actually interact with. It retains the model’s general reasoning and coding abilities but comes without the full suite of cybersecurity features. It can help you write code, analyze data, and handle complex reasoning tasks, but it won’t help you find zero-day exploits in production software.

Advertisement

Pricing for the new models is set at $10 per million input tokens and $50 per million output tokens. That’s double what previous Claude Opus versions cost.

On the benchmark side, the model posted a +31 point improvement on the USAMO 2026 math competition compared to its predecessors.

Why crypto should be paying attention

Mythos 5’s capabilities reportedly extend to infrastructure-level risks: key management systems, oracle implementations, and the underlying blockchain mechanics that most audits don’t touch. For an industry that has lost billions to exploits over the past several years, the existence of an AI that can systematically probe these deeper layers is both a threat and an opportunity. The threat is obvious: if a model like this leaks, gets replicated, or falls into the wrong hands, the attack surface for DeFi protocols expands dramatically. The opportunity is that legitimate security teams with access to Mythos 5 could identify vulnerabilities before attackers do, potentially closing gaps that manual audits routinely miss.

What this means for investors

For crypto investors, the traditional checklist — smart contract audit from a reputable firm, bug bounty program, multisig wallet — may soon be insufficient. If AI models can probe oracle designs, key management architectures, and cross-chain bridge implementations at machine speed, the protocols that survive will be the ones that have access to equivalent defensive tools.

Anthropic is pricing Mythos-era models at a significant premium over previous Claude Opus versions, suggesting confidence that the performance gap over competitors justifies the cost. For enterprise users in crypto, particularly exchanges, custodians, and large DeFi protocols, the question becomes whether the security benefits of restricted-access AI tools justify the expense relative to traditional audit and monitoring approaches.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.