Microsoft's Satya Nadella criticises Anthropic's Fable AI restrictions

Microsoft CEO Satya Nadella has criticised Anthropic's approach to content restrictions in its flagship Fable AI model, arguing that creative AI tools should not be subject to what he described as excessive editorial control.

Nadella questions Fable's response limits

Speaking during an internal meeting with Microsoft engineers working on the company's Copilot AI software, Nadella criticised the way Anthropic's premium Fable model refuses certain user requests.

"If you use Fable, when it refuses for any random thing, it just is like, when was the last time you had a creation tool that was so editorially controlled?" Nadella said, according to remarks obtained by CNBC.

"It doesn't make sense."

Microsoft declined to comment on the remarks, while Anthropic did not immediately respond to CNBC's request for comment.

Anthropic tightened safeguards after relaunch

Anthropic's support documentation notes that Fable may redirect some requests involving large-scale AI model development and other sensitive topics to older model versions.

When it launched Fable 5 in June, Anthropic said it had reduced false positives that unnecessarily blocked harmless prompts. However, after temporarily restricting access to comply with US export controls, the company restored the model on July 1 with stricter safeguards that would block a slightly higher proportion of benign requests.

Microsoft pushes lower-cost AI models

Nadella's comments come as businesses increasingly explore smaller, more cost-efficient AI models instead of relying solely on expensive frontier models from leading AI labs.

On Thursday, Chinese startup Moonshot AI unveiled an open-source model it claimed outperformed recent releases from Anthropic and OpenAI.

Nadella has also argued that organisations should be able to build customised AI models using their own data without sharing it with external model developers.

Microsoft's Foundry platform currently gives developers access to more than 11,000 AI models, including offerings from Anthropic and OpenAI.

"It can't be that there are only two companies in the world with token capital, and everybody else is renting it," Nadella told engineers. "It makes no economic sense."

Criticism comes despite Microsoft-Anthropic partnership

The remarks are notable because Anthropic remains both a Microsoft partner and customer.

Microsoft invested $5 billion in Anthropic last year, while the AI startup agreed to spend $30 billion on Microsoft's Azure cloud platform. Microsoft also uses Anthropic's models to power features within its enterprise AI assistant, Copilot Cowork.

Anthropic's Claude Code has also gained popularity among developers and non-technical users for software development tasks.

Microsoft continues expanding its AI strategy

Nadella also defended Microsoft's decision to merge its consumer and enterprise Copilot products under a single leadership structure, saying the company should have unified the products from the beginning.

"The unification is something we should have done maybe day one," he said.

Microsoft announced in April that it had surpassed 20 million paid Copilot seats for enterprise customers, representing roughly 4% of its cloud-based Office customer base.

The company has also expanded its own portfolio of in-house AI models while broadening Foundry's catalogue, reflecting a strategy of offering customers a wide range of AI options rather than relying exclusively on a single model provider.