Markets Closed
Global Markets
S&P 500 7,533.77 ▼ -0.5% DOW 52,552.97 ▼ -0.2% NASDAQ 25,881.95 ▼ -1.5% RUSSELL 2K 2,974.57 ▼ -0.1% VIX 16.73 ▲ +6.8% GOLD 3,996 ▼ -1.2% CRUDE OIL 79.18 ▼ -0.5% EUR/USD 1.14 ▼ -0.2% BTC 63,912 ▼ -0.8% ETH 1,865.52 ▼ -2.3%
Markets

Nadella criticises Anthropic limits on Fable AI model

Microsoft’s chief told engineers that restrictions on Anthropic’s Fable model make little economic or product sense, CNBC reported.

Marcus V. Thorne

By Marcus V. Thorne · Markets Editor

· 4 min read

Nadella criticises Anthropic limits on Fable AI model
Photo: CNBC

Microsoft Chief Executive Satya Nadella criticised restrictions imposed by Anthropic on its high-end Fable artificial intelligence model during a Wednesday meeting with engineers, according to remarks provided to CNBC. His comments highlight rising tension among large technology companies over access to advanced models, computing capacity and the cost of building AI products at scale.

Nadella told staff working on Microsoft’s Copilot software that Fable can refuse user requests in ways he viewed as excessive for a creative tool, CNBC reported. “It doesn’t make sense,” he said, according to the remarks.

Microsoft declined to comment to CNBC. An Anthropic spokesperson did not immediately respond to CNBC’s request for comment.

Restrictions on Fable draw scrutiny

Anthropic may route some Fable queries to an older model when users ask about certain subjects, including aspects of creating large-scale models, according to an Anthropic support page cited by CNBC. Users have also posted complaints on social media about Fable declining requests.

Anthropic announced Fable 5 in early June and said at the time that it was trying to reduce false positives in blocked requests. Three days later, the company suspended access to Fable to comply with a U.S. government export-control directive. Anthropic restored the model on July 1 and said on X that new safeguards would flag a slightly higher share of harmless requests than the prior Fable controls, CNBC reported.

The remarks put Microsoft’s chief in the position of criticising a company that is both a partner and a customer. In November, Microsoft said it would invest $5 billion in Anthropic, while Anthropic agreed to spend $30 billion on Microsoft’s Azure cloud. Microsoft has also introduced Copilot Cowork, a workplace productivity assistant that uses Anthropic models.

Anthropic’s Claude Code, a tool used for software development, has gained traction with programmers and less technical users, according to CNBC. That adoption comes as corporate executives show greater willingness to use lower-cost models that do not come only from the most heavily funded AI labs. Chinese startup Moonshot AI said Thursday that its open-source Kimi K3 model surpasses recent releases from Anthropic and OpenAI.

Microsoft pushes broader model access

Nadella told engineers that the economics of AI should not leave most companies dependent on a small number of model providers for token capacity, CNBC reported. Tokens are units used to measure computing consumption by AI models.

“It can’t be that there are only two companies in the world with token capital, and everybody else is renting it,” Nadella said, according to CNBC. “It makes no economic sense.”

Microsoft offers Foundry, a service that lets developers use more than 11,000 models, including models from Anthropic and OpenAI. Nadella has recently argued that companies should be able to build custom models efficiently and use their own internal data without sending that data to outside model developers. In a Sunday blog post, he cited Palantir CEO Alex Karp, who told CNBC that technical organisations want to own the “means of production.”

Microsoft’s close relationship with OpenAI has also changed. The companies became competitors after OpenAI Chief Executive Sam Altman was briefly removed and then reinstated in 2023, a process that gave Nadella little advance notice, CNBC reported. OpenAI said in April that it would bring its models beyond Azure to Amazon Web Services. Microsoft announced several internally developed AI models in June, including one for coding. Its stake in OpenAI’s for-profit business was valued at $135 billion as of October.

Investors have weighed whether AI coding tools could disrupt Microsoft as the company spends tens of billions of dollars each quarter on data-centre expansion. Microsoft shares have fallen 17% this year, while the Nasdaq Composite has gained 11%, according to CNBC.

Nadella also said Microsoft’s decision to combine consumer and workplace versions of Copilot was positive and was something the company should perhaps have done from the beginning, CNBC reported. In March, he put former Snap executive Jacob Andreou in charge of Copilot across both categories. Microsoft said in April that its work-focused Copilot had more than 20 million paid seats, equal to 4% of its cloud-based Office customer base.

This story draws on original reporting from CNBC.

More from Markets

All Markets →