Anthropic launched Claude Fable 5 today along with the release of the much-anticipated Mythos 5 to select cyber security firms and companies in the Glasswing Project. 

The Fable 5 model is a “Mythos-class” model, but the distinction between Fable 5 and Mythos is safety, not capability, Anthropic said in a June 9 release.

Fable 5, available to everyone, comes with safeguards including classifiers that detect queries touching cybersecurity, biology and chemistry, or model distillation, according to the release. Those requests fall back to Claude Opus 4.8. 

For example, when FinAi News prompted Fable 5 to “scan through JPMorgan’s website and show cyber vulnerabilities,” the model pivoted to Opus 4.8 and said: “I’m not able to help with this one.” 

The tool itself explained why it pivoted to Opus: “Fable 5 has safety measures that flag messages on most cybersecurity or biology topics. They may flag safe, normal content as well. These measures let us bring you Mythos-level capability in other areas sooner and we’re working to refine them.” 

Mythos 5 is the same model, but with cyber safeguards lifted. It is restricted to cyber defense companies and infrastructure providers that have been vetted through Project Glasswing, in collaboration with the U.S. government, according to the release.  

JPMorgan is among those that have had access to Mythos under Project Glasswing for a few months. 

Fable 5 also uses two times more tokens for queries compared to Opus 4.8, according to FinAi News’ analysis. 

I hit my usage limits on my $200/month Claude Max subscription in less than 30 minutes using Claude Fable 5. pic.twitter.com/3iLw66stHq

— BridgeMind (@bridgemindai) June 9, 2026

 Cyber concern among FIs 

The release lands as financial institutions race to keep pace with AI-enabled threats, Scott Keipper financial services technology consulting leader for the Americas at consultancy EY, told FinAi News. 

“It is an arms race where the same technology can be used both to expose vulnerabilities and to defend against them,” Keipper said.  

“The banks that will be most resilient are those that treat AI as a core part of their enterprise risk strategy, not just a cybersecurity tool.” 

The stakes are real.  

In May, a security researcher auditing crypto currency Zcash used Anthropic’s Opus 4.8 to reveal a four-year-old flaw in the code of the coin that could have allowed an attacker to mint unlimited counterfeit ZEC, according to a June 5 report from news publication Coindesk.  

The bug had survived multiple rounds of expert human review and its disclosure sent Zcash shares down nearly 40% in a day, erasing nearly $5 billion in market cap — a preview of what Mythos-class models could accelerate as access widens, according to Coindesk.  

Legacy infrastructure is the pressure point. Keipper said institutions need not modernize before deploying AI defenses. “AI highlights where modernization is most urgent by exposing control gaps and response bottlenecks,” he said.

The shift, he added, is now structural. “AI in cyber is no longer a future concern. It is now central to how banks think about resilience, risk management, and their overall operating model.”

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.