Anthropic Releases Most Powerful AI Model Yet, with Enhanced Safety Features

A new milestone in artificial intelligence has been reached by Anthropic, a company that specializes in developing advanced language models. The firm has released Claude Fable 5, its most capable publicly available AI model to date, alongside Claude Mythos 5, a restricted version accessible only to trusted cybersecurity and scientific research partners.

The dual release marks an important shift in the development of AI systems, as companies like Anthropic are now experimenting with separate safety measures for their most advanced models. According to Gabe Goodhart, Chief Architect of AI Open Innovation at IBM, this move was inevitable given the rapid progress being made by other labs in the field.

Anthropic's Fable 5 model has demonstrated exceptional performance on a range of benchmarks, including software engineering, knowledge work, vision, and scientific research. The company claims that its new safety features have been designed to prevent misuse in areas such as cybersecurity, biology, and chemistry.

The release of Fable 5 also highlights the growing concern about the potential risks associated with advanced AI systems. Regulators, researchers, and security experts have raised questions about how to prevent these tools from being used for malicious purposes, such as cyberattacks or other harmful activities.

To address this issue, Anthropic has implemented a layered safety architecture that includes AI-powered classifiers designed to detect risky requests before they reach the underlying model. When users submit prompts involving sensitive topics, the system routes those requests to a less capable model instead of allowing them to access the full capabilities of Fable 5.

According to Goodhart, this approach places an unusual emphasis on the risks posed by the underlying model and acknowledges that even with safety measures in place, there is still potential for harm. He notes that Anthropic's decision to pair its most capable public model with a separate layer of safeguards may signal a broader change in how AI companies package their advanced systems.

The pricing and distribution of frontier AI models are also being reevaluated by the industry. As demand for Fable 5 grows, Anthropic plans to shift from subscription-based access to a usage-credit model at least temporarily. This move could have significant implications for users who require high levels of intelligence in their applications.

Goodhart suggests that this change may indicate a trend towards holding back the best models from subscription plans and offering them instead through more expensive, tiered pricing structures. He notes that this would result in higher costs for users seeking advanced AI capabilities.