Nadella Publicly Criticizes Anthropic's Fable AI Restrictions to Microsoft Staff

Microsoft CEO Satya Nadella told employees this week that Anthropic's restrictions on its Fable AI model don't make sense, according to CNBC, which obtained a copy of his remarks to engineers working on Microsoft's Copilot software.

"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 CNBC. "It doesn't make sense."

This is a critical comment directed at a company Microsoft has poured billions into. Microsoft announced a $5 billion investment in Anthropic in November, and Anthropic agreed to spend $30 billion on Microsoft's Azure cloud. Anthropic's Claude Code tool has become popular with developers, and Microsoft built its Copilot Cowork product partly on Anthropic's models. Microsoft declined to comment further; Anthropic did not respond to CNBC's request for comment.

The Fable rollout has been a mess

Anthropics introduced Fable 5 in early June, saying at the time it was trying to cut down on false-positive refusals, according to CNBC. Three days later, the company cut off Fable access entirely to comply with a U.S. government export control directive. Anthropic restored the model on July 1, but admitted the new safeguards would "flag a slightly higher fraction of harmless requests than the previous Fable safeguards," per Anthropic's own statement cited by CNBC.

Anthropic loosened the rules, got ordered to shut the model down by the federal government, then brought it back with rules that are tighter than before. Users have been calling out the rejections on social media, and Anthropic's own support page acknowledges the model sometimes falls back to an older version when it flags certain requests, according to CNBC.

What actually happened in June

The shutdown itself was abrupt and, by multiple accounts, poorly explained. According to the Times of India, citing reporting that included internal employee chats later viewed by The New York Times, the White House ordered Anthropic's most powerful models, Fable 5 and Mythos 5, taken offline with less than 90 minutes' notice on a Friday in June. The stated justification was national security, but Anthropic's roughly 3,000 employees reportedly weren't given specifics.

Internal messages described a shifting rationale: first the concern was foreign companies gaining access to the models, then a vulnerability inside them, then both, according to the Times of India's account of the employee chats. The order blocked all foreign nationals worldwide, along with green-card holders working inside the U.S., from using the models. This restriction was so broad that Anthropic reportedly had to pull access for everyone to comply. Executives were said to be on the phone with government officials within 15 minutes of the shutdown call asking for a clearer reason, and reportedly didn't get one.

A company forced to disable its flagship product with 90 minutes' notice and given a shifting explanation rather than a fixed one has a real complaint about how the process was handled. Whether the underlying security concern was valid is a separate question from whether the government communicated it competently. No public charges, indictments, or formal findings have been disclosed alongside the order, based on available reporting, and the specific vulnerability or foreign-access risk that triggered it has not been detailed publicly.

Export controls on advanced AI models aren't unprecedented, and a national security agency moving fast and tight-lipped on a classified or sensitive matter isn't automatically evidence of bad faith. The concern from Anthropic's side, that the explanation kept changing and left engineers guessing, is worth taking seriously. It's also possible the government's story evolved because officials themselves were still nailing down the scope of the problem, not because they were making it up as they went.

Microsoft's real motive isn't hidden

Nadella's criticism isn't happening in a vacuum. CNBC reports that Microsoft has been pushing companies to build cost-efficient, custom AI models using their own internal data rather than depending entirely on outside labs like Anthropic and OpenAI. In a blog post over the weekend, Nadella cited Palantir CEO Alex Karp's comment that technical organizations "want to know they own the means of production," according to CNBC.

Microsoft also runs the Foundry service, which offers developers access to more than 11,000 models, including ones from Anthropic and OpenAI, per CNBC. So Microsoft has a direct financial interest in customers not feeling locked into Anthropic's ecosystem, restrictions and all.

The timing matters too. Chinese startup Moonshot AI released an open-source model, Kimi K3, that it claims scores higher on knowledge-work benchmarks than Anthropic's Claude Opus 4.8 Max, trailing only Claude Fable 5 Max and GPT-5.6 Sol Max, according to figures reported by futunn citing GDPval-AA v2 data. Microsoft investors have already been nervous. Microsoft shares are down 17% so far this year while the Nasdaq Composite is up 11%, according to CNBC, partly on fears that fast-improving AI coding tools could undercut Microsoft's own business.

Nadella's specific complaint about over-blocked requests may be valid. But his public criticism of a company Microsoft just