The launch of Claude Fable 5 on June 9 has stirred up significant debate within the cybersecurity community. Experts are voicing concerns about its practical applications, particularly in identifying and addressing vulnerabilities in critical software before malicious actors can exploit them. This model is Anthropic’s first public offering from the Mythos-class of artificial intelligence systems and is equipped with safety classifiers that divert high-risk inquiries to the previous version, Claude Opus 4.8. Topics such as cybersecurity, biology, and chemistry prompt this fallback, effectively rendering Fable 5 less effective for cybersecurity professionals who need precise tools for vulnerability analysis.
What does Fable offer and why is it significant?
Fable 5 possesses the same foundational capabilities as the earlier Claude Mythos 5, which demonstrated impressive skills in pinpointing software flaws. During a testing phase in April 2026, the Mythos-class models successfully identified over 23,000 critical vulnerabilities within major codebases. Anthropic responded to these findings by creating a public-facing version that retains the general intelligence while cleaning up the riskier aspects. The company claims that over 95% of Fable 5 interactions do not require the fallback to Opus 4.8, yet this statement may overlook the concerns of specialists who depend on functionality in critical areas like cybersecurity.
The dilemma of dual access
Anthropic appears to be establishing a two-tiered access structure. General users have access to Fable, while vetted professionals and organizations may eventually receive access to the complete Mythos suite, which keeps the unrestricted capabilities intact. The pricing for Fable 5 is set at $10 per million input tokens and $50 per million output tokens, which is approximately double the cost associated with Opus 4.8. This presents a troublesome choice for users who may find that they are paying more for less functionality within certain professional workflows. Currently, access to Fable is restricted to paid subscribers only.
What does this mean for AI and security?
Anthropic’s Project Glasswing has shown the potential of these models to uncover vulnerabilities at scale, uncovering over 23,000 critical vulnerabilities during earlier tests. Unfortunately, the classifiers that flag all cybersecurity-related inquiries do not differentiate between the efforts of legitimate security researchers and those of malicious hackers. This could lead to genuine security work being obstructed while astute bad actors likely figure out ways to circumvent these restrictions.
If Anthropic's cautious strategy forces security professionals to seek alternatives, the company risks alienating a valuable segment of its user base without achieving tangible improvements in security outcomes.
As the methods surrounding AI and cybersecurity continue to evolve, the implications of Fable 5 remain a pivotal point of discussion. Experts must navigate the balance between safety and functionality in these advanced systems to ensure that defenders are adequately supported against evolving threats.