- Anthropic has launched Claude Fable 5, based on technology from its restricted Mythos model.
- The original Mythos system was withheld over concerns it could help hackers identify cyber vulnerabilities.
- Fable 5 is more powerful than Anthropic's previous public model but comes with stricter guardrails and a higher price tag.
Just two months after warning that one of its most advanced AI systems could pose cybersecurity risks, Anthropic has released a public version built from the same underlying technology, marking a notable shift in how the company balances capability and safety.
The artificial intelligence company announced Claude Fable 5 on Tuesday, bringing some of the power behind its restricted Mythos model to a wider audience. Anthropic had previously declined to release Mythos broadly, arguing that the system was capable enough to help malicious actors identify weaknesses in computer networks and potentially aid cyberattacks.
The launch places Anthropic at the centre of a growing debate within the AI industry: how much capability is too much capability when powerful models can be used by both defenders and attackers.
From Security Fears To Public Release
When Anthropic unveiled Mythos in April, the company took the unusual step of limiting access to a small group of organisations responsible for critical infrastructure. Rather than making the model available to consumers or businesses, it was provided to selected groups so they could identify vulnerabilities in their own systems before criminals had the chance to exploit them.
The decision attracted attention from cybersecurity experts, policymakers and business leaders, many of whom questioned how companies should handle AI systems capable of accelerating cyber research.
Now Anthropic believes it has found a middle ground.
According to the company, Fable 5 includes additional safeguards designed to block responses related to areas considered particularly sensitive, including cybersecurity and biology. When users attempt to access information that Anthropic classifies as too risky, requests are instead routed to Claude Opus 4.8, a model released last month that was developed without the same security concerns attached to Mythos.
"That is the core of why we have been able to release Fable," Anthropic's head of product management Dianne Penn reportedly said.
The approach, however, may create trade-offs. While the restrictions could make it harder for hackers to misuse the technology, they may also limit its usefulness for cybersecurity professionals seeking to strengthen digital defences.
Bigger Performance, Bigger Costs
Anthropic says benchmark testing carried out by AI evaluation firm Vals AI found that Fable 5 outperformed every publicly available AI model tested by the company. According to those results, the model scored around 5 per cent higher overall than Claude Opus 4.8.
The stronger performance comes with a significantly higher price.
Anthropic will charge around £7.40 ($10) for every million input tokens and roughly £37 ($50) for every million output tokens. That is about double the cost of Opus 4.8.
Tokens are the units AI companies use to measure usage. A software developer working on a large coding project can consume millions of tokens within a relatively short period, making pricing an increasingly important factor for businesses adopting AI tools at scale.
The launch also comes as competition among leading AI firms intensifies and the costs of building advanced models continue to rise.
Anthropic recently signed a deal to lease data centre capacity from Elon Musk's xAI, a move that reportedly costs the company around £925 million ($1.25bn) each month. The agreement highlights the enormous infrastructure demands facing AI companies as they race to develop more capable systems.
The release of Fable 5 follows a busy period for the sector. Both Anthropic and OpenAI have recently disclosed confidential filings for potential US stock market listings, while investors continue to scrutinise how AI companies plan to fund the growing costs associated with developing and running increasingly sophisticated models.
For Anthropic, Fable 5 represents more than just another product launch. It is an attempt to answer a question that many AI developers are still grappling with: how do you make powerful technology widely available without creating new risks in the process?
