Anthropic Releases Two AI Models — One for Partners, a Safer One for You

Anthropic released two new versions of its advanced AI on Tuesday, making one available to the public while keeping its most powerful capabilities exclusive to select partners. The company launched Claude Fable 5, its first "Mythos-class" model for general use, alongside Claude Mythos 5, a more potent version reserved for trusted organizations. This dual-pronged release strategy marks a new chapter in the industry's struggle to balance cutting-edge performance with self-professed safety commitments.

The public model, Fable 5, is being positioned as the most powerful AI Anthropic has ever made widely available. According to The Verge, the company claims it “shows exceptional performance in software engineering, knowledge work, and vision,” especially on long and complex tasks. Yet, this power comes with explicit limitations. As reported by TechCrunch and Wired, Anthropic has implemented guardrails that prevent the model from generating responses in high-risk domains, specifically citing cybersecurity and biology as areas where its capabilities have been deliberately curtailed.

A Tale of Two Models: Mythos and Fable

The core of the announcement is the bifurcation of Anthropic's top-tier technology. While the public gets Fable 5, a select group of vetted partners gains access to Mythos 5. VentureBeat reports that this continues the work from Anthropic's restricted cybersecurity program, known as Project Glasswing, where its most advanced models were tested in a controlled environment. The consensus across reports from The Guardian to BBC Business is that Fable 5 is a deliberately throttled version of the technology that previously caused a stir among government and finance leaders due to its advanced capabilities.

This isn't just a simple feature flag. Anthropic is creating a clear distinction between what paying enterprise partners can access and what the general public can experiment with. The strategy allows Anthropic to commercialize its leading-edge research and demonstrate its progress in the AI race, while simultaneously attempting to create a liability shield by preventing the public version from being used for what it deems dangerous applications. Wired notes the company's explicit statement that Fable 5 “can’t be used for cyberattacks,” a direct nod to the safeguards baked into the public model.

The Rationale Behind the Guardrails

The restrictions are the defining feature of this public release. For months, Anthropic has signaled that its most powerful models were too dangerous for broad access, with The Guardian noting that access had been restricted over cybersecurity concerns. The creation of Fable 5 is the company's answer to that self-imposed dilemma. By building in non-negotiable restrictions, Anthropic believes it can safely release a highly capable model without enabling malicious use cases.

This approach is a direct result of the company's public commitment to AI safety. It also reflects a structural reality: as AI models become more adept at coding and understanding scientific principles, their potential for dual-use applications—both beneficial and harmful—grows exponentially. Blocking queries related to offensive cybersecurity or the synthesis of biological agents is a brute-force, but currently necessary, method of risk mitigation. It’s a practical, if inelegant, solution to a problem the entire industry faces.

Beyond the Hype: A Calculated Commercial Strategy

This tiered release is more than just a safety protocol; it’s a shrewd business strategy. It establishes a premium tier of service—access to the unrestricted Mythos 5—that can be sold to high-value corporate and government clients who pass a vetting process. This creates an enterprise-grade offering that is clearly differentiated from the free or low-cost public version. The pattern indicates a move to monetize the apex of its technology with customers who have both the budget and, presumably, the trustworthiness to handle it.

Together, these reports point to a new model for AI deployment that sits between the “move fast and break things” ethos of early tech and the completely closed-off, lab-only approach. Anthropic is attempting to have it both ways: capture market share and developer interest with a powerful public model, while reserving its true state-of-the-art for a select, high-paying clientele. This suggests the future of AI access may not be a simple binary of open versus closed, but a complex spectrum of tiered, permissioned access. The term “public release” now requires an asterisk, as the most powerful tools remain behind a corporate-controlled gate.