Just two months after unveiling an AI model so powerful that it raised alarms among cybersecurity experts and government officials, AI company Anthropic is now making a toned-down version available to the public.
Back in April, Anthropic introduced a new artificial intelligence system called Claude Mythos. The company said it was too risky to release on a global scale as it could help hackers discover ways to break into computer networks.
The warning sparked concern among business leaders and policymakers who feared the technology could open the door to a new wave of cyberattacks. Now, Anthropic says it has found a way to safely release a version of that technology.
What is Claude Fable 5
On Tuesday, the company announced Claude Fable 5, a new AI model built using some of the capabilities behind Mythos but with strict safeguards in place. According to Anthropic, Fable includes additional guardrails that prevent it from providing responses related to sensitive areas such as cybersecurity, biology and other fields that could potentially be misused.
The company believes these restrictions will make it much harder for hackers to use the model to attack computer systems. However, those same limitations could also make it more difficult for cybersecurity professionals and businesses to use the AI to strengthen their own defences.
According to the firm, when users ask questions in areas considered “too risky,” most of those requests will instead be handled by Claude Opus 4.8, a model released last month that was specifically designed to avoid the security concerns associated with Mythos. “That is the core of why we have been able to release Fable,” said Anthropic’s head of product management, Dianne Penn.
According to benchmark testing conducted by AI evaluation company Vals AI, Fable performs better overall than any publicly available AI model. The tests found that Fable’s overall performance is 5% higher than Claude Opus 4.8.
Why Mythos was kept under lock and key
Rather than making Mythos available to everyone, Anthropic initially provided access to around 40 organisations responsible for maintaining critical computer infrastructure.
The goal was to help these organisations identify and fix security loopholes before cybercriminals could find a way to break in. This month, Anthropic said it plans to expand access, giving another 150 organisations the opportunity to use Mythos.
Even now, Mythos remains available only to a limited number of groups and does not contain the same guardrails that have been added to Fable.
Fable 5 comes at a higher cost
The new model is significantly more expensive than Anthropic’s previous public release.
Fable 5 costs $10 for every million input tokens and $50 for every million output tokens. That is roughly twice the price of Opus 4.8.
Tokens are the units AI companies use to measure usage. A programmer working on an intensive coding project can easily use up one million tokens within a few hours or even less.
Anthropic recently signed a deal to lease a data centre from Elon Musk’s xAI, which is part of SpaceX. The arrangement reportedly costs the company $1.25 billion every month.
AI industry heads toward IPO wave
The launch of Fable 5 comes during a period of intense excitement around artificial intelligence.
Both Anthropic and its biggest rival, OpenAI, announced over the past week that they had filed plans for initial public offerings.
Meanwhile, SpaceX is expected to make its stock market debut on Friday in a listing that many analysts believe could break records.