Anthropic has made Fable 5 — a new member of its Mythos product family — broadly available to the public, while continuing to limit the model’s abilities for certain sensitive use cases. The Mythos line was introduced by the company in April as its most advanced AI platform, but access was initially confined to a small set of partner organizations due to cybersecurity concerns.
What Fable 5 is designed for
According to Anthropic, Fable 5 excels at writing and debugging software code, answering complex research questions, and analysing visual content. The company emphasises that the public Fable 5 does not always use the full capability set of the Mythos models.
Who is affected by the restrictions and how filtering works
Anthropic says most queries related to cybersecurity, biology, or chemistry are automatically routed to the Opus 4.8 model, which the company released at the end of May and describes as a less capable system. Attempts to probe the model’s internals or to replicate its technology are similarly handled by the lower‑capability model.
The rationale is that Mythos models are highly effective at uncovering vulnerabilities in critical infrastructure — for example, in banking systems, power grids and other essential networks — which could pose risks if exploited.
Full functionality reserved for partners
Alongside the public Fable 5 offering, Anthropic still provides the full‑function Claude Mythos 5 to organisations that previously participated in the Mythos programme. That group includes cybersecurity partners involved in the Project Glasswing initiative. As of early June, the programme’s participant list had grown to roughly 200 organisations across more than 15 countries, and Anthropic says the network may expand further.
Testing, red‑teaming and bug bounty
Anthropic engaged external experts to stress‑test the system: red‑teaming specialists logged more than 1,000 work hours trying to bypass built‑in safeguards. The company also launched a bug bounty programme offering financial rewards for found vulnerabilities. Anthropic reports that no one has fully lifted the Mythos models’ restrictions.
Governmental and political context
When Project Glasswing launched, some critics accused Anthropic of overstating the technology’s risks to gain market attention. Since then, multiple organisations that tested the system have corroborated Mythos’ capabilities. The US government examined the model on security grounds, and developments led the White House to create a cooperation framework that enables leading AI developers’ strongest models to be tested ahead of public release. The rollout of Mythos 5 has proceeded in coordination with US government actors.
Anthropic also clashed with the Trump administration after refusing to lift restrictions related to mass surveillance and autonomous lethal weapon systems; as a result, the Pentagon ended contracts with the company. That was a sensitive blow given that Anthropic’s tools previously held certain defence security clearances.
Pricing and business situation
Fable 5 is materially more expensive than the company’s earlier models: Anthropic charges $10 per one million input tokens and $50 per one million output tokens, roughly double the pricing of Opus 4.8. Tokens are the billing units for AI services; an intensive programming session can consume a million tokens in a few hours.
Although Anthropic’s revenues are growing rapidly, the company remains unprofitable and spends heavily on compute capacity. It recently began renting data‑centre capacity from Elon Musk’s xAI for $1.25 billion per month; xAI operates within the SpaceX group.
Why this matters
The public release of Fable 5 opens new possibilities for business and research users, but Anthropic’s cautious access policy highlights that governance and risk management for advanced language models remain central concerns. Decisions about model capabilities and access will shape how such systems spread across industry and government applications.
