Satya Nadella says Anthropic’s Claude Fable restrictions ‘don’t make sense’ | Technology News
4 min readNew DelhiJul 17, 2026 09:02 AM Microsoft CEO, Satya Nadella, has criticised Anthropic over its limits on user requests to its high-end artificial intelligence (AI) model Claude Fable. The CEO, on Wednesday, said that the limits ‘don’t make sense’.
According to a CNBC report, Nadella told software engineers working on Microsoft’s Copilot AI that Anthropic’s restrictions reminded him of “the last time we had a creation tool that was so editorially controlled.”
Nadella’s comments appeared to target Anthropic’s policy that restricts users from accessing outputs from its latest AI model, Claude Fable, to train or improve competing AI models. This practice is known as model dillation. Reportedly, when end users ask Fable about aspects needed to create large-scale models, Anthropic returned responses from an older model, as per its support page.
Some users have taken to social media to express their displeasure. A user, Derya Unutmaz whose bio claims to be immunolog, shared a post on X with a screenshot showing how the model shifted to Opus 4.8 version when he asked about a critical question.
Simply put, Anthropic does not want competitors to query Claude Fable thousands or millions of times, collect its responses, and use them to build rival AI models. According to the company, this is effectively copying years of expensive research and development. Anthropic has earlier alleged that some companies attempted such large-scale dillation.
Why does Anthropic switch models?
According to Anthropic, Claude Fable 5 may automatically switch to the older Claude Opus 4.8 model if it detects requests in high-risk areas such as offensive cybersecurity, biology, chemry, AI model dillation, or certain frontier AI development tasks. This is because Fable 5 is far more capable than previous public models, and Anthropic wants to reduce the risk of misuse while still making it widely available. The safety system scans the entire conversation, including uploaded files, memory, and web results, not just the latest prompt. Users can disable automatic switching, but blocked requests will then require manual edits or a model change.
Nadella’s comments come at a time when many companies are increasingly turning to more cost-effective AI models from emerging players instead of depending solely on the largest, best-funded AI labs, particularly for software development and enterprise applications.Story continues below this ad
The remarks also follow rising competition in the AI industry. On Thursday, Chinese startup Moonshot AI unveiled an open-source model that it claims outperforms some of the latest offerings from Anthropic and OpenAI on key benchmarks.
His criticism is particularly notable as Anthropic is both a strategic partner and a major Microsoft customer. Anthropic’s Claude Code has become a popular AI coding assant among developers and non-technical users. Last year, Microsoft announced a $5 billion investment in Anthropic, while the startup committed to spending $30 billion on Microsoft’s Azure cloud platform. Microsoft has also integrated Anthropic’s models into its enterprise offerings, including the Copilot Cowork productivity assant.
According to the CNBC report, Nadella also argued that companies should develop custom AI models cost-effectively and use their internal data without sharing it with companies that build foundation models. He said enterprises should own their data and AI infrastructure instead of relying solely on a single AI model provider.
