Nadella criticizes Anthropic’s Fable limits as Microsoft pushes AI control

Microsoft CEO Satya Nadella told a Wednesday meeting of engineers working on the company’s Copilot AI software that Anthropic’s limits on requests submitted to its high-end Fable model “don’t make sense,” per remarks provided to CNBC.

Nadella zeroed in on the model’s rejection pattern. “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?” he told the engineers.

Microsoft declined to comment. Anthropic did not immediately respond.

What Fable actually rejects and why?

Anthropic launched Fable 5 in early June and said at the time it was trying to reduce false positives on blocked requests. Three days after the launch, the US government’s export control directive forced Anthropic to disable Fable and Mythos entirely.

When Anthropic restored the model on July 1, the company said “The new safeguards will flag a slightly higher fraction of harmless requests than the previous Fable safeguards.” Anthropic’s own support pages note that when users ask Fable about certain topics, including some aspects of creating large-scale models, the platform may route responses through an older model version.

The pattern has surfaced in enterprise complaints on social media. Microsoft’s own legal team restricted internal employee use of Fable 5 back in June over Anthropic’s 30-day retention policy for prompts and outputs, while continuing to sell the model to external Copilot and Azure Foundry customers.

Nadella pushes back on renting token capacity

The Fable criticism was the sharpest quote in Nadella’s remarks, but not the load-bearing one.

Nadella told the engineers:

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

The argument extends the case Nadella has been building publicly for a month. As Cryptopolitan earlier reported, Nadella used a June Wall Street Journal interview and a July 12 X blog post to argue that a small group of dominant frontier model providers cannot sustainably capture the value from every enterprise’s proprietary workflow data.

His Sunday post invoked Palantir CEO Alex Karp, who told CNBC that technical organizations “want to know they own the means of production.”

Microsoft’s Foundry service currently offers over 11,000 models, including those from Anthropic and OpenAI, and the company announced in June that it was building its own in-house coding model.

Microsoft’s Anthropic bet grows more complicated

Nadella’s remarks land at an awkward moment for the Microsoft-Anthropic relationship. In November, Microsoft committed a $5 billion investment in Anthropic, while Anthropic agreed to spend $30 billion on Microsoft’s Azure cloud.

Earlier this year, Microsoft unveiled Copilot Cowork, a business productivity assistant that draws on Anthropic’s models. Anthropic’s Claude Code has meanwhile become popular among Microsoft’s own engineering customer base. Meanwhile, Microsoft’s OpenAI relationship has been fraying since the abrupt 2023 CEO ousting and reinstatement of Sam Altman with little notice to Nadella.

OpenAI said in April it would bring its models to Amazon Web Services. Microsoft’s stake in OpenAI’s for-profit business was worth $135 billion as of October, and Microsoft shares have fallen 17% year-to-date while the Nasdaq Composite has gained 11%.

Investors are pricing the risk that Microsoft’s Azure economics erode if enterprise customers find cheaper alternatives. Meanwhile, Chinese startup Moonshot AI announced an open-source model on Thursday that it said outperforms recent Anthropic and OpenAI releases.

If you're reading this, you’re already ahead. Stay there with our newsletter.