Anthropic is making its most advanced AI model available to the general public for the first time, but with significant safety guardrails in place. The move comes just days after the company warned that AI systems are advancing so rapidly they could soon achieve autonomous self-improvement without human oversight.
This release marks a significant shift in how AI companies balance public access with safety concerns. As competition intensifies between major AI labs, Anthropic’s approach of releasing a restricted version of its most powerful technology could set a new industry standard for deploying advanced models.
Claude Fable 5 launches with built-in safety limits
On Tuesday, Anthropic launched Claude Fable 5, the first publicly available version of its Mythos model. The company says Fable 5 excels in several key areas:
- Software engineering tasks
- Knowledge work applications
- Vision-related capabilities
However, the model comes with hard safety limits. In high-risk areas like cybersecurity, biology, chemistry, and distillation processes, Fable 5 blocks responses and automatically falls back to the less powerful Claude Opus 4.8 model.
Limited rollout reflects security concerns
The path to public release has been cautious. Anthropic first launched Mythos as a preview in April, limiting access to a handful of partners due to cybersecurity concerns. Last week, the company expanded access to hundreds of organizations across 15 countries, focusing on those managing critical infrastructure.
Now, anyone can access a version of this technology through Anthropic’s Claude API and consumption-based Enterprise plans. The company is rolling out subscription access in stages:
- Through June 22: Fable 5 included in Pro, Max, Team, and seat-based Enterprise plans at no extra cost
- From June 23: Model removed from standard plans, requiring usage credits
- Future: Plans to restore as standard subscription feature when possible
For organizations already approved for advanced model access, Anthropic is also deploying Mythos 5, a new version of the underlying technology.
Extensive testing reveals no universal vulnerabilities
Concerned about potential misuse, Anthropic conducted extensive security testing before the public release. The company stress-tested its safety classifiers with jailbreak attempts and ran an external bug bounty program.
“Internally, we ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing. We then worked with external red-teaming orgs which also failed to find universal jailbreaks,” the company stated.
Despite this testing, Anthropic acknowledges that novel attacks could still be possible. As a result, the company now requires 30-day data retention on all traffic, even for enterprises that previously had zero-retention agreements. The company says it won’t use this data for training, only to defend against attacks and reduce false positives.
Early performance data shows promise
According to Anthropic, cases where Fable 5 must defer to the older Opus 4.8 model are rare. Early data shows at least 95% of Fable sessions run entirely on the newer model’s responses.
Third-party testing has shown strong results. Analytics company Hex reported that Fable was the first model to achieve 90% on its core analytics benchmark of complex, long-running tasks. “On the hardest questions, it shows strong judgment and attention to nuance,” Hex noted.
Other early adopters have highlighted specific strengths:
- Base44 praised Fable’s ability to create full applications in single attempts
- Genspark found it performed significantly better on UI design and game coding tasks
- Rakuten values the model’s ability to reflect on and validate its own work
Premium pricing reflects advanced capabilities
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens – double the cost of Opus 4.8. This premium pricing might naturally limit widespread adoption.
The high cost comes as many enterprises are becoming more critical of AI expenses after exceeding budgets or receiving unexpectedly large bills. Advanced models can worsen these issues by splitting single requests into multiple tasks.
However, some companies see the value proposition. Rakuten noted that “the extra thinking pays for itself” when it comes to enabling highly autonomous operations.
Anthropic expects demand for Fable 5 to be very high and difficult to predict, suggesting the company anticipates strong adoption despite the premium pricing and safety restrictions.
