Anthropic on Tuesday unfurled Claude Fable 5, the first time the IPO-bound artificial intelligence (AI) company is making its all-powerful Mythos technology available to the public albeit bound by strict safety guardrails and controversial data-retention policies.
Initially unveiled as a restricted preview in April, the underlying Mythos model was held back from the public due to severe cybersecurity concerns. Early testing revealed the AI could autonomously discover vulnerabilities across every major operating system and web browser.
To mitigate these risks, Anthropic initiated Project Glasswing — a massive security collaboration involving tech titans such as Amazon Web Services, Apple Inc., Google, Microsoft Corp., and JPMorgan Chase — to identify and patch software bugs before a wider rollout. (Later Tuesday, AWS launched Claude Fable 5 on both Amazon Bedrock and the Claude Platform on AWS.)
While Claude Fable 5 boasts unprecedented capabilities in software engineering, vision, and complex knowledge work, Anthropic has implemented a hard “brake pedal.”
If a user queries the model on high-risk topics — specifically cybersecurity, biology, chemistry, or distillation — the system will block the request and seamlessly defer to the company’s next-most-capable model, Claude Opus 4.8. Early data indicates this safety triggers are conservative but rare, with 95% of user sessions running entirely on Fable 5.
Wary of the implications of an unaligned frontier system, Anthropic put Fable 5 through more than 1,000 hours of internal and external red-teaming jailbreak attempts and reported zero universal vulnerabilities.
However, acknowledging that novel attack vectors remain possible, Anthropic is instituting a 30-day data retention policy on all Fable 5 and Mythos 5 traffic. This mandate applies even to enterprise clients with prior zero-retention agreements. While Anthropic emphasizes that this data will not be used for training, industry experts note the policy could set a significant precedent, effectively tying frontier AI access to mandatory data monitoring.
“The differentiator isn’t raw power, it’s that they wrestled it into something safe enough to release widely. The value is in the scale,” said Chris Boehm, field chief technology officer of Zero Networks. “When only 200 organizations have a capability, it barely moves the needle for the industry. When everyone can point it at their own code and infrastructure, defenders finally get to operate at the same speed as the people attacking them. That’s worthwhile, assuming the safeguards hold up, which is the thing I’ll be watching in the model card.”
Early enterprise feedback indicates that Fable 5 represents a significant leap in autonomous operations. Stripe Inc. reportedly compressed months of codebase migration work into a single day. Hex said Fable 5 was the first AI to score 90% on its complex analytics benchmark. And Rakuten Group Inc. highlighted the model’s ability to autonomously reflect on and validate its own work.
This level of autonomy carries a premium. Fable 5 and the enterprise-exclusive Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens — twice the cost of Opus 4.8.
However, some cybersecurity experts still see threats.
“Anthropic’s primary mitigation appears to be routing sensitive cybersecurity and biological queries to lower-capability models.,” said Alfredo Hickman, chief information security officer at Kai. “While this reduces risk, it introduces a new attack surface: the routing layer itself. If an attacker can find prompts that avoid triggering the classifier while still extracting portions of Mythos-class reasoning, the safeguard effectively becomes a detection problem rather than a capability problem. Security practitioners know that detection systems rarely achieve perfect coverage.”
The rollout arrives just one week after Anthropic filed confidential IPO paperwork with the SEC, joining OpenAI and SpaceX in a race toward what Wall Street expects to be some of the largest public offerings in history.
The launch also follows Anthropic’s urgent public plea for global AI labs to coordinate safety boundaries. The company warned that frontier systems are advancing so rapidly they could soon achieve recursive self-improvement, autonomously upgrading their own software without human intervention. Fable 5 represents Anthropic’s high-stakes gamble that it can safely commercialize the frontier of autonomous AI while keeping its foot firmly on the safety brake.
“As new models emerge faster than traditional security cycles can absorb, the organizations best positioned will be those that pair innovation with resilience,” Anthony Grieco, chief security and trust officer at Cisco Systems Inc., said in an email. “That means continuing to invest in the fundamentals that never go out of date: patching, MFA, segmentation, and zero trust. AI will raise the ceiling for what defenders can do, but security resilience remains the foundation that determines whether those gains translate into real protection.”
