Anthropic has made its most powerful artificial intelligence (AI) model, Mythos, available to the general public for the first time, while restricting its ability to carry out cybersecurity-related tasks.

 

The company on Tuesday launched Claude Fable 5, the first publicly available version of its Mythos model. According to Anthropic, Fable 5 delivers strong performance in software engineering, knowledge work, and vision-related tasks. However, the model has been equipped with hard safety limits in high-risk domains, including cybersecurity, biology, chemistry and model distillation, Bloomberg reported.

 

When users attempt tasks in these restricted areas, Fable 5 blocks the request and instead falls back to Claude Opus 4.8. Anthropic said these safeguards are designed to reduce the risk of misuse, making it harder for malicious actors to use the model to attack computer networks. Anthropic is also releasing the same model, without some of the safeguards, as a new version of Mythos called Mythos 5. It will be available to the groups that can use the cyber-capable model through an initiative called Project Glasswing.

 
 


“We wanted to make sure for non-cyber use cases, we really prioritised safely releasing Fable as soon as possible,” Dianne Penn, head of project management for Anthropic’s research and labs, said. “That’s why we’re bringing this part of the Fable piece first while we continue to work on the general cyber use cases”, he added.

 


Designed for coding and complex professional tasks

 


Anthropic said Fable 5 has been designed to improve performance in coding and other professional applications, particularly tasks that require solving complex problems over extended periods compared with earlier models.

 


The model currently leads all publicly available AI systems in overall performance, according to benchmark tests conducted by Vals AI, a company that tracks the capabilities of leading AI models.Fable’s overall performance was 5 per cent higher than that of Claude Opus 4.8, reported The New York Times

 


The enhanced capabilities however come at a higher cost, with Fable 5 priced at roughly twice the cost of Opus 4.8.

 


Anthropic conducted safety testings

 


According to Bloomberg, Anthropic conducted an external bug bounty programme to assess whether users could bypass Fable 5’s safety mechanisms.

 


The company said external researchers and red-team testers spent more than 1,000 hours attempting to jailbreak the model and circumvent its restrictions. Anthropic reported that the exercise did not uncover any universal jailbreak methods capable of consistently bypassing the safeguards.

 


Mythos access expanded to India and nations

 

The launch follows Anthropic’s recent expansion of access to the Mythos model through Project Glasswing. Last week, the company extended access to 150 organisations across more than 15 countries, including India.

 


The expansion also covered members of the Five Eyes intelligence alliance: Canada, Australia and New Zealand. Other countries granted access include France, Germany, Italy, Switzerland, the Netherlands, Spain, Belgium, Sweden, Japan and South Korea, the Financial Times reported, citing people familiar with the matter. The North Atlantic Treaty Organisation (Nato), headquartered in Brussels, has also been given access to the technology.

 


Anthropic said it plans to continue expanding the number of organisations eligible to use the cyber-capable version of Mythos through Project Glasswing.



Source link

YouTube
Instagram
WhatsApp