Fable 5 redirects cybersecurity, biology, and chemistry queries to Claude Opus 4.8.
Unrestricted Mythos 5 operates only under Project Glasswing with the US government.
Anthropic launched on June 9, 2026 Claude Fable 5, the first model of the Mythos class available for general use. According to the company, the model surpasses in capacity all those it has made publicly available to date, with frontier performance in software engineering, scientific research and complex knowledge work.
However, the firm launched it with mechanisms that block cybersecurity queriesbiology, chemistry and model distillation, redirecting those answers to Claude Opus 4.8.
The reason for the blockage refers directly to what Anthropic documented in the weeks leading up to the launch. Through Project Glasswing, the restricted program with which Anthropic distributed Mythos Preview since Aprilthe company and its approximately 50 partners found more than 10,000 vulnerabilities of high or critical severity in what Anthropic describes as the most systemically important software globally.
Evaluation participants include Cloudflare, Mozilla, Microsoft, Oracle and Palo Alto Networks, as well as banks and critical infrastructure institutions.
Cloudflare identified 2,000 errors in critical path systems — 400 of them high or critical severity.with a false positive rate that the company team considered better than that of human auditors. Mozilla, for its part, found and fixed 271 vulnerabilities in Firefox 150, more than ten times what it detected in Firefox 148 with Claude Opus 4.6.
The UK AI Safety Institute noted that Mythos Preview was the first model in solve your two simulations of cyberattacks from start to finish autonomously.
That same offensive power is what Anthropic decided to restrict in Fable 5. Without safeguards, the model’s capabilities in areas such as cybersecurity could be used improperly to cause serious damage, the company notes in its announcement.
The company warns that filters are conservatively calibrated and will sometimes block harmless requests.
Unrestricted Mythos – Defenders Only
Fable 5 differs from Mythos 5, its unrestricted counterpart. Mythos 5 has the same capabilities as Fable 5, but with safeguards lifted in some areas. It will initially be deployed through Project Glasswing, in collaboration with the US government, as an update to Mythos Preview.
Anthropic notes that it has the most advanced cybersecurity capabilities of any model in the world. Although, compared to its main competitors, Anthropic maintains one of the most restrictive positions on the market. While OpenAI and Google DeepMind have progressively relaxed some filters to increase the usefulness of their models, Anthropic continues to prioritize security, even at the cost of limiting functionality.
This “constitutional” strategy has been both praised by regulators and criticized by developers seeking greater flexibility.
The context in which Fable 5 reaches the public is that of a documented escalation of attacks against cryptocurrency infrastructure with AI assistance. According to Maximiliano Carjuzaa, co-founder of Money On Chain, close to 100% of the DeFi attacks recorded in the previous months involved AI to some extent. Ledger CTO Charles Guillemet warned in April that AI is lowering the barrier to entry for attackers.
In that same month, the industry recorded at least 34 hacks with losses close to USD 635 million, equivalent to almost 80% of the total stolen so far in 2026.
For the cryptocurrency ecosystem, the arrival of Fable 5 to the public poses a specific tension: the same filters that Anthropic considers sufficient for general use have not been verified independently in the face of systematic evasion attempts.
Fable 5 blocks cybersecurity queries, but the unrestricted version—Mythos 5—already operates in the hands of government and infrastructure organizations, with capabilities that Project Glasswing logs demonstrate as unprecedented in autonomously detecting critical vulnerabilities.