Anthropic launches powerful AI Mythos, but blocks cybersecurity features
NewsSunday, 21 June 2026 at 00:29
Anthropic has released a public version of its controversial AI model Mythos—this time without the cybersecurity features that sparked global backlash.
According to Reuters, the company is also launching a new model, Claude Fable 5, designed to deliver Mythos-level power without the risks tied to hunting software vulnerabilities.
The move is a major pivot for Anthropic as it tries to balance cutting-edge AI with guardrails—while its rivalry with OpenAI heats up.
Why Mythos dominated the spotlight
Earlier this year, Anthropic quietly rolled out Mythos to a small set of organizations through its Glasswing program. The company says the model uncovered thousands of software flaws, including issues missed by traditional security tools.
Those results drew both praise and alarm across the cybersecurity world. An AI that can find vulnerabilities could just as easily be used to map attack paths—or exploit them.
As a result, Anthropic initially limited access to roughly 200 organizations, including parts of the U.S. government.
Claude Fable 5 becomes the public flagship
Rather than fully opening Mythos, Anthropic is making Claude Fable 5 its most powerful publicly available model.
The company says Fable 5 excels at:
- Software development
- Data analysis
- Complex business workflows
- Research and knowledge work
According to Dianne Penn, Head of Product Management, Research & Labs at Anthropic, the model includes built-in safeguards to prevent misuse.
If users attempt to get help finding vulnerabilities in software packages or source code, the system refuses. In those cases, the platform falls back to a less capable model, Claude Opus 4.8, which lacks advanced cyber capabilities.
Safety becomes a selling point
Fable 5’s debut underscores how AI companies are increasingly positioning safety as a core product feature.
Where earlier AI generations were judged mainly on raw performance, the focus is shifting to control, access, and risk reduction.
Anthropic says it has extensively tested the system to block users from bypassing restrictions via jailbreaks or prompt manipulation.
That emphasis is notable as regulators press the industry on the societal risks of ever more powerful AI.
The OpenAI showdown enters a new phase
The timing is striking: Anthropic and OpenAI are converging on valuation, market share, and technical capability.
Reuters reports Anthropic has reached an estimated valuation of about $965 billion. That would put it ahead of OpenAI, though private valuations can swing with funding rounds and investor sentiment.
Both companies are seen as prime candidates for future IPOs in the AI sector.
A broader deployment of Mythos-era tech could further strengthen Anthropic’s standing with enterprises and governments seeking advanced AI solutions.
Cutting the cost per task
Anthropic is pitching Fable 5 as not just more capable—but more efficient.
Early users say it consumes fewer tokens to complete the same tasks, which can lower total job costs even if the model sits in a higher price tier.
Official pricing is:
- $10 per million input tokens
- $50 per million output tokens
That puts Fable 5 firmly in the premium bracket of the generative AI market.
What this means for the AI landscape
Fable 5 signals the next phase of AI: success isn’t just about intelligence—it’s about what a model is intentionally prevented from doing.
More AI vendors are adopting a tiered approach: the most powerful systems go to a trusted few, while the broader market gets versions with built-in limits.
For governments and large enterprises, that can be appealing: access to advanced AI without unrestricted cyber capabilities. But it also raises questions over who gets the most capable systems—and on what terms.
With Claude Fable 5, Anthropic is betting that safety, trust, and controlled access will matter as much as sheer model performance.
Latest comments
Loading