Nadella Slams Anthropic's Fable Restrictions: 'Makes No Sense'
Miles Bennett
Microsoft CEO Satya Nadella openly criticized partner Anthropic's flagship model Fable for randomly refusing requests, calling the behavior 'illogical' at an internal staff meeting. The rare public split signals growing tension between Microsoft's multi-model AI strategy and its suppliers' product choices.
What exactly did Nadella say?
Per an internal transcript obtained by CNBC, Nadella told Copilot engineers: Fable randomly refuses all kinds of requests — "What creative tool have you ever used that was editorially controlled like that? It doesn't make sense."
This means → Microsoft internally views Fable's restriction mechanism as a product-experience flaw, not a safety feature.
Anthropic did not respond to a request for comment; Microsoft also declined — both sides chose silence.
What is Fable's restriction mechanism?
When users ask Fable about certain topics — including large-scale model training — the system may return older-version responses instead of answering directly. Some users have publicly criticized this on social media.
Anthropic acknowledged the issue when it launched Fable 5 in early June, saying it was working to reduce false refusals. But just three days after launch, Fable access was suspended to comply with a U.S. government export-control directive.
When service resumed on July 1, Anthropic admitted the new safety measures would slightly raise the false-refusal rate for harmless requests. In plain terms = tighter safety comes at the cost of more legitimate queries being blocked — exactly the point Nadella attacked.
How deep does this partnership run?
Last November Microsoft announced a $5 billion investment in Anthropic; Anthropic in turn committed to $30 billion in Azure cloud spending.
This year Microsoft also launched Copilot Cowork, an enterprise productivity assistant built on Anthropic's models.
This means → Nadella is not criticizing an outsider. He is publicly flagging a product flaw in a partner Microsoft has bet billions on — which makes the disagreement far heavier.
What is Nadella's bigger play?
At the same meeting Nadella said: "It is not possible that only two companies have compute capital and everyone else is renting it — that makes no economic sense."
In a recent blog post he cited Palantir CEO Alex Karp's view, stressing that enterprises should own "the means of production."
This reflects Nadella's strategic throughline: Microsoft wants to be the AI infrastructure platform, not hostage to any single model supplier's product decisions.
What does Microsoft's AI alliance map look like now?
Microsoft's Foundry service offers developers over 11,000 models, including those from Anthropic and OpenAI.
The relationship with OpenAI has cooled since the 2023 Altman departure crisis; the two are now competitors.
Microsoft shares are down 17% year-to-date, while the Nasdaq Composite is up 11% — the market is not pricing Microsoft's AI strategy optimistically.
How should we read this?
Nadella chose to criticize Anthropic at an internal meeting rather than privately — that itself is a deliberate signal.
This means → the key question going forward is whether Microsoft will further diversify its model procurement, reducing dependence on both Anthropic and OpenAI.
In plain terms = the $5 billion investment still stands, the $30 billion commitment still stands, but the temperature of the partnership has clearly dropped a notch.
Content is for reference only, not financial advice.