Anthropic has unveiled its latest generation of AI systems, the Mythos-class models, marking a significant advancement in artificial intelligence capabilities. This launch introduces two distinct versions: Claude Fable 5, accessible to the general public, and Claude Mythos 5, a more tightly controlled variant reserved for select organizations and security partners.
What are Mythos-Class Models?
Mythos-class models represent Anthropic's newest category of AI systems, succeeding the company's earlier Opus and Sonnet families. These models are engineered to tackle complex, long-running tasks demanding sustained reasoning. Anthropic highlights their particular strength in areas such as software engineering, scientific research, advanced problem-solving, and knowledge-intensive workflows where AI must maintain context and reasoning over extended periods.
The Mythos family initially appeared earlier this year through a limited-access preview. However, Anthropic delayed a wider release due to concerns regarding some of its inherent capabilities.
Claude Fable 5: The Public-Facing AI
Claude Fable 5 is the first Mythos-class model made available to the broader public. Anthropic describes it as its most capable generally accessible AI model to date, designed for developers, researchers, and businesses requiring AI assistance for coding, analysis, research, and complex agentic workflows.
A key distinguishing feature of Fable 5 compared to previous Claude models is the integration of new safety systems. Anthropic states that Fable 5 can detect specific categories of high-risk requests and apply additional safeguards before generating responses. The model underwent extensive internal testing, external red-teaming exercises, and jailbreak evaluations prior to its release.
Claude Mythos 5: Restricted Access for Advanced Capabilities
Claude Mythos 5 shares the same underlying architecture and core capabilities as Fable 5. However, the crucial difference lies in its lack of some additional restrictions and safety controls that are built into Fable 5. Anthropic previously indicated that earlier Mythos models demonstrated exceptionally strong cybersecurity capabilities, including the ability to identify and exploit software vulnerabilities. These findings prompted the company to limit access while it thoroughly evaluated potential risks.
Consequently, Mythos 5 remains under strict control and is not being broadly released. Access is managed through trusted-access programs and Project Glasswing, an initiative focused on securing critical software infrastructure.
Why Two Versions? Anthropic's Tiered Approach
Anthropic's decision to release two nearly identical versions with vastly different access rules addresses a significant challenge for AI developers: how to deploy increasingly powerful models without making potentially dangerous capabilities widely accessible. Rather than permanently restricting the technology, Anthropic has adopted a tiered strategy.
- Fable 5 serves as the public-facing model, equipped with additional safeguards.
- Mythos 5 is reserved for vetted organizations that may require access to the model's full capabilities for specialized research, advanced cybersecurity, and critical infrastructure-related work.
This approach allows Anthropic to make its latest technology available while maintaining tighter oversight of higher-risk use cases.
Safety Measures in Fable 5
One notable safety addition in Fable 5 is a routing system designed for sensitive requests. If users attempt tasks involving advanced cybersecurity, biology, or chemistry topics, Fable 5 can either refuse the request or automatically transfer the interaction to the less capable Claude Opus 4.8 model. Anthropic asserts this method helps reduce access to potentially dangerous capabilities without hindering everyday use.
The company also conducted adversarial testing, bug bounty programs, and external security reviews before launch. According to Anthropic, testers were unable to identify a universal jailbreak that could reliably bypass the model's safeguards.
Who Can Access Each Model?
Accessing Claude Fable 5
Fable 5 is immediately available to paying Claude customers, including subscribers to Pro, Max, Team, and Enterprise plans. Anthropic is also integrating the model into various developer ecosystems and third-party platforms, such as GitHub Copilot and cloud services that provide access to Claude models.
Accessing Claude Mythos 5
Mythos 5 remains restricted to a small group of approved organizations. Access is primarily managed through trusted-access programs and Project Glasswing. Previous versions of Mythos were exclusively available to vetted governments, security researchers, infrastructure operators, and select organizations. While Anthropic has indicated that access will gradually expand, no timeline for a broader public release has been announced.