Anthropic has announced the launch of Claude Fable 5, its latest flagship AI model for general availability, alongside Claude Mythos 5, a more capable version that will initially be restricted to cybersecurity organisations and selected research partners.
The company describes Fable 5 as its most capable publicly available model to date, with improvements across software engineering, knowledge work, vision, scientific research and long-context reasoning. According to Anthropic, the model demonstrates stronger performance on a range of benchmarks compared with its previous-generation systems.
Focus on enterprise and complex workloads
Anthropic said the model has been designed to handle long-running and complex tasks more effectively, enabling greater autonomy across software development, analytical workflows and research-orientated use cases.
Early testing cited by the company indicates improvements in software engineering productivity, including large-scale code migration and codebase management tasks. Anthropic also highlighted gains in document analysis, financial reasoning, visual understanding and long-context memory capabilities.
The company noted that Fable 5 can maintain context across very large datasets and extended interactions, making it suitable for enterprise workflows that require sustained reasoning and memory retention.
Restricted release of Claude Mythos 5
Alongside Fable 5, Anthropic introduced Claude Mythos 5, a version of the same underlying model that operates with fewer restrictions in specific domains, particularly cybersecurity and scientific research.
Initially, access to Mythos 5 will be limited through Project Glasswing, a programme developed in collaboration with US government agencies and selected cybersecurity organisations. Anthropic plans to expand access through a broader trusted access programme in the future.
The company stated that Mythos 5 exhibits advanced capabilities in cybersecurity, scientific reasoning and life sciences research. Internal testing highlighted applications ranging from software security analysis to drug discovery and genomics research.
New safeguards for frontier AI models
Anthropic’s announcement places significant emphasis on safety controls designed to address risks associated with increasingly capable AI systems.
The company said that Fable 5 incorporates a new layer of safety classifiers that automatically redirect certain categories of requests to a less capable model, Claude Opus 4.8. Areas subject to additional scrutiny include offensive cybersecurity, biology, chemistry and AI model distillation activities.
According to Anthropic, these safeguards have been implemented to reduce the risk of misuse while enabling broader access to advanced AI capabilities. The company acknowledged that some legitimate requests may be affected by the controls, describing the approach as intentionally conservative during the initial rollout phase.
The safeguards reflect growing industry concerns around the potential misuse of frontier AI models in areas such as cyberattacks, biological research and the replication of advanced AI systems.
Changes to data retention policies
Anthropic also announced changes to its data handling practices for Mythos-class models and future systems of similar capability.
Under the new policy, business customer interactions with these models will be retained for 30 days for safety and security monitoring purposes. The company stated that the retained data will not be used for model training and will be subject to additional privacy controls and deletion requirements.
The move reflects a broader trend among AI developers to strengthen monitoring and governance mechanisms as model capabilities continue to advance.
Growing debate around frontier AI governance
The launch comes amid increasing industry focus on balancing access to powerful AI systems with safeguards designed to mitigate security and societal risks.
As AI models demonstrate stronger capabilities in areas such as software development, scientific research and autonomous task execution, technology providers are introducing more sophisticated governance frameworks to control access, monitor usage and prevent misuse.
Anthropic’s dual-release strategy, offering a broadly available model with additional safeguards alongside a restricted-access version for trusted organisations, highlights one emerging approach to managing the deployment of frontier AI capabilities while maintaining access for enterprise users and researchers.