Anthropic Releases Claude Mythos Model Fable 5 to the Public

Thailand AI News: AI Firm Opens Public Access to Previously Restricted Frontier Model

Anthropic has officially released Claude Fable 5, the first publicly accessible version of its highly advanced Mythos artificial intelligence model, marking a major milestone in the evolution of frontier AI systems. The announcement comes after months of restricted testing and growing speculation surrounding Mythos, a model that had previously been available only to a select group of organizations because of concerns about its extraordinary capabilities and potential security implications.

The company describes Fable 5 as its most capable publicly available AI system to date, excelling in software engineering, knowledge work, visual understanding, and complex analytical reasoning. Unlike the original Mythos model, however, Fable 5 incorporates extensive safeguards designed to prevent misuse in sensitive areas such as cybersecurity, biology, medicine, chemistry, and AI model distillation.

Anthropic says that when users attempt to access capabilities within these high-risk domains, the system automatically defers to the less powerful Claude Opus 4.8 model. In releasing the technology to a global audience, this Thailand AI News report highlights how Anthropic is attempting to balance innovation with safety at a time when concerns about advanced AI capabilities are intensifying across governments, corporations, and research institutions worldwide.

From Limited Testing to Global Availability

When Anthropic first introduced Mythos in April, access was tightly restricted to a handful of trusted organizations. The company justified the decision by pointing to concerns that the model’s advanced reasoning abilities could potentially be leveraged to identify vulnerabilities in computer systems or assist in sophisticated cyber operations.

Last week, Anthropic expanded access to several hundred organizations across 15 countries, focusing primarily on entities responsible for critical infrastructure and cybersecurity operations. The launch of Claude Fable 5 now extends a version of the technology to the broader market through Anthropic’s Claude API and consumption-based enterprise offerings.

For a limited period through June 22, users on Pro, Max, Team, and seat-based Enterprise plans will receive access at no additional cost. After that date, Anthropic plans to move Fable 5 to a usage-based pricing structure before eventually restoring it as a standard subscription feature.

Anthropic Introduces Mythos 5 for Approved Organizations

Alongside the public rollout of Fable 5, Anthropic has also unveiled Mythos 5, a new version of the advanced model that will remain available only to approved organizations.

Unlike Fable 5, Mythos 5 does not include the same restrictions on cybersecurity and biological research capabilities. Access remains limited to trusted organizations operating in specialized sectors where advanced AI capabilities can be used for defensive and infrastructure-related purposes.

Anthropic stated that organizations participating in the preview program have already used Mythos to uncover more than 10,000 critical security vulnerabilities across their systems. The company expects access to expand gradually through a broader trusted access initiative in the coming months.

Extensive Safety Testing Before Public Release

Recognizing the concerns surrounding a Mythos-class model, Anthropic conducted extensive testing before making Fable 5 available to the public.

According to the company, internal researchers organized an external bug bounty program that generated more than 1,000 hours of testing focused on bypassing the model’s safeguards. Anthropic claims that no universal jailbreak capable of consistently defeating the model’s protections was discovered during these exercises.

The company also collaborated with external red-teaming organizations to identify weaknesses and evaluate potential attack vectors. While Anthropic acknowledged that novel attacks may still emerge, executives argue that the testing process provided confidence that the model could be released responsibly.

New Data Retention Policy May Set Industry Precedent

One of the most significant aspects of the launch is Anthropic’s decision to implement mandatory 30-day data retention for all traffic involving Fable 5 and Mythos 5.

Even enterprise customers that previously operated under zero-retention agreements will be subject to the new policy. Anthropic stressed that retained information will not be used to train future models. Instead, the company says the data will be used exclusively to detect sophisticated attacks, identify jailbreak attempts, improve safety systems, and reduce false positives.

Industry observers suggest this policy could become a template for other AI companies as increasingly powerful models enter the market. Access to frontier AI systems may increasingly come with enhanced monitoring requirements intended to protect against misuse.

Strong Early Performance Results

Early evaluations suggest that Fable 5 is among the strongest AI systems currently available.

Analytics company Hex reported that Fable became the first model to achieve a score of 90 percent on its benchmark for complex analytical reasoning tasks. The company praised the model’s ability to handle nuanced questions and long-running analytical workflows.

Other organizations reported similarly impressive results. Base44 highlighted the model’s ability to generate complete software applications from a single prompt, while AI workspace provider Genspark stated that Fable outperformed competing models in tasks involving user-interface design, coding, and autonomous tool usage.

Rakuten also praised the model’s capacity for self-evaluation, noting that Fable can review and validate its own work before delivering results, a capability that could significantly enhance autonomous operations.

Growing Debate Over Powerful AI Systems

The release comes amid increasing debate about the pace of AI development and the potential risks posed by rapidly advancing systems.

Anthropic recently urged major AI laboratories to establish mechanisms that would allow the industry to slow development if necessary. Company co-founder Jack Clark argued that while AI companies currently possess a powerful “gas pedal” accelerating innovation, there remains no equivalent “brake pedal” capable of slowing progress should risks become unacceptable.

These concerns have become more pronounced as researchers discuss the possibility of recursive self-improvement, a scenario in which AI systems become capable of autonomously improving their own capabilities without direct human intervention.

A Defining Moment for Anthropic

The public release of Claude Fable 5 represents a defining moment not only for Anthropic but also for the broader AI industry. The company is attempting to demonstrate that highly capable frontier AI systems can be made available to the public while maintaining meaningful safeguards against misuse. Whether those protections prove sufficient will be closely watched by regulators, enterprises, researchers, and competitors alike. As the race toward increasingly powerful AI accelerates, Fable 5 may ultimately be remembered as one of the first major tests of whether advanced AI can be deployed responsibly at scale while preserving both innovation and public trust.

For more details, visit:

https://www.anthropic.com/news/claude-fable-5-mythos-5

For the latest AI models or platforms, keep on logging to Thailand AI News.