Satya Nadella questions Anthropic's Claude Fable limits, says restrictions 'don't make sense'

Microsoft CEO Satya Nadella has publicly questioned Anthropic's approach to limiting access to its flagship artificial intelligence model, Claude Fable, arguing that some of the company's safeguards are difficult to justify. His remarks arrive at a time when competition among AI developers is intensifying and businesses are increasingly looking for greater flexibility in how they build and deploy AI systems.

Speaking to Microsoft engineers working on the company's Copilot AI products, Nadella criticised Anthropic's restrictions on certain categories of user requests. According to a CNBC report, he said the limits "don't make sense" and compared the company's approach to "the last time we had a creation tool that was so editorially controlled."

More from Tech

The comments are notable because Anthropic is not only a major player in the AI race but also maintains close commercial ties with Microsoft through cloud infrastructure and enterprise AI offerings.

Restrictions aimed at preventing AI model copying

At the centre of the debate is Anthropic's policy designed to prevent users from extracting knowledge that could help build competing AI systems. The company has introduced safeguards that restrict Claude Fable from answering certain prompts related to AI model development, particularly those that could facilitate model distillation—a technique in which developers repeatedly query a powerful model and use its responses to train another system.

Rather than providing answers from Claude Fable in such situations, Anthropic may instead route the request to an earlier-generation model with lower capabilities. The company says this reduces the risk of its latest research being used to accelerate rival AI projects without permission.

Anthropic has previously claimed that some organisations attempted large-scale distillation of its models, making protective measures necessary. According to its support documentation, the system evaluates the entire conversation—including uploaded documents, memory and web results—before deciding whether a request falls into a restricted category.

The automated switch is not limited to AI-related prompts. Anthropic says it may also activate for requests involving offensive cybersecurity, biology, chemistry and other sensitive frontier AI topics. Users can disable automatic switching, although requests that violate the platform's safety policies may still require manual changes before they can proceed.

Some users have expressed frustration with the behaviour. Among them was immunologist Derya Unutmaz, who shared screenshots on X showing Claude Fable switching to the older Claude Opus 4.8 model after being asked what he described as a critical technical question.

AI competition moves beyond model size

Nadella's criticism reflects a broader shift taking place across the AI industry. Rather than relying exclusively on the largest foundation models, many enterprises are increasingly exploring smaller, specialised systems that can be customised using their own data while reducing operational costs.

According to CNBC, Nadella argued that organisations should be able to develop AI models efficiently without handing over valuable proprietary information to foundation model providers. He also suggested that businesses should retain ownership of both their data and the infrastructure powering their AI deployments instead of becoming dependent on a single external model.

His comments come as competition among AI companies continues to accelerate. On Thursday, Chinese startup Moonshot AI introduced a new open-source model that it claims surpasses some of the latest systems from Anthropic and OpenAI on several benchmark tests, underscoring the growing pressure on established AI labs.

A delicate relationship despite public criticism

The remarks are particularly striking given Microsoft's relationship with Anthropic. Despite questioning the company's AI policies, Microsoft remains one of Anthropic's important strategic partners.

Claude Code has become one of the more widely adopted AI-assisted coding tools among software developers, while Microsoft has integrated Anthropic's models into several enterprise services, including its Copilot Cowork productivity platform.

The two companies also share significant commercial interests. Last year, Microsoft announced a $5 billion investment in Anthropic, while the AI startup committed to spending $30 billion on Microsoft's Azure cloud platform.

Nadella's comments therefore highlight a growing tension emerging across the AI industry. Even companies that collaborate closely on infrastructure and commercial partnerships are increasingly diverging on how powerful AI systems should balance openness, commercial protection and user access. As AI models become more capable—and more valuable—the debate over where to draw those boundaries is only expected to intensify.