- Key insight: The U.S. federal government gave Anthropic approval to once again start selling a version of its Mythos AI model called Fable 5.
- What's at stake: The company had been forced to shut down Mythos because it was too good at finding cybersecurity gaps, and then it was forced to table Fable over export controls.
- Forward look: The back and forth between a major frontier model provider and the federal government has some industry observers wondering if companies will rethink their reliance on a small number of big AI vendors that seem to be at the mercy of government officials' whims.
Anthropic has been given the federal government's green light to start selling Fable 5, its Mythos-with-guardrails model, again.
Fable 5 will be available starting Wednesday to users globally of Claude Platform, Claude.ai, Claude Code, and Claude Cowork, and users of other platforms will get access this month.
The company was forced to shut down Mythos 5, its powerful model that finds cybersecurity vulnerabilities at unprecedented speeds, and the guardrailed Fable version of it two weeks ago due to a U.S. government export control that required it to restrict access to foreign nationals, for security reasons that weren't provided at the time.
In a blog, Anthropic explained that the June 12 export control directive came after the government became aware of a report in which Amazon researchers had found a method of bypassing Fable 5's safeguards: prompting it so that it identified a number of software vulnerabilities. Anthropic's testing found that other, less capable models could also be jailbroken using the same method.
The company has been working with the government, Amazon and other partners to fix this potential security risk. It's created several "classifiers," which are "smaller automated AI systems that, during an interaction, detect when the model is asked to perform a potentially harmful cybersecurity task (or produces potentially harmful outputs)," Anthropic said in a post on its website
If a classifier blocks a request, the user will be sent to an older Anthropic model, Opus 4.8.
"Thanks? I guess," commented Simon Taylor, CEO of FintechBrainfood, on the fallback to Opus.
Anthropic has also restored access to Mythos 5 to a small group of organizations that are members of Project Glasswing, a coalition of mainly tech companies, and JPMorganChase, that were
The back and forth between a major frontier model provider and the federal government (on top of the Pentagon dubbing Anthropic a supply chain risk earlier this year) has some industry observers wondering if companies will rethink their reliance on a small number of big AI vendors that seem to be at the mercy of government officials' whims.
"Most enterprises will eventually use multiple AI models, not because one model is best at everything, but because different tasks require trade-offs between cost, speed and reasoning," said Sachin Puri, CEO of Network Solutions, a website registrar. "The real innovation isn't asking customers to choose between various models, but intelligent orchestration that chooses for them. The best platforms will adopt a customer-centric approach that automatically routes simple tasks to fast, efficient models and reserves frontier reasoning models for the problems that truly require them."
And some believe Anthropic brought government scrutiny upon itself with claims about how powerful its models are.
"AI companies spent years teaching the public and government to treat frontier models as unknowable, quasi-mythic objects: things coming for your livelihood, to be feared and worshipped rather than understood," Andrew Atiya, founder and CEO of Knowhow, wrote on LinkedIn. "Now Anthropic is discovering what happens when that narrative turns back on them."
