Microsoft CEO Satya Nadella has publicly pushed back against the content restrictions built into Anthropic's high-end Claude Fable model, telling employees the safeguards "don't make sense" for what is meant to be a creative tool. Speaking to engineers on Microsoft's Copilot team, Nadella said: "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's criticism extended past refused prompts to Anthropic's broader practice of restricting how its models can be used to train or improve competing AI systems, often called "distillation."
He reportedly called it hypocritical for AI labs to defend fair use of internet data for their own training while restricting rivals from learning off their outputs, and argued Microsoft wants AI to stay broadly accessible rather than concentrated among a handful of companies, warning against enterprises becoming "compute sharecroppers" dependent on a few model providers. Microsoft currently offers access to over 11,000 AI models through its Azure AI Foundry platform, including Anthropic's, OpenAI's, and its own.
Fable's Rocky Launch
The comments come after a rocky period for Fable. In June, Anthropic released Claude Fable 5 together with Claude Mythos 5, introducing additional safety measures in areas such as biology and cybersecurity, while stating it aimed to reduce the number of benign requests that were mistakenly blocked. A few days later, both models were taken offline—not because of content moderation issues, but in response to a U.S. Department of Commerce export control order that required a license for use by any foreign national. Access was reinstated on July 1, with Anthropic acknowledging that the updated safeguards would probably cause a slightly higher proportion of harmless requests to be flagged than before.
The friction is notable given how intertwined the two companies are: Microsoft has invested roughly $5 billion in Anthropic and relies heavily on Claude Code internally, while Anthropic has committed tens of billions of dollars to run and train its models on Azure. Microsoft remains OpenAI's largest backer too, spreading its AI bets across multiple frontier labs rather than one.
The exchange underscores a growing tension between safety-driven restrictions and enterprise demand for flexible AI tools. For Anthropic, guardrails protect against misuse and years of costly research being copied; for Microsoft, tightly controlled models risk pushing developers toward more permissive alternatives, including a rising wave of open-source systems from Chinese labs. Whether this becomes a lasting rift or a passing remark, it points to a fault line likely to resurface as frontier AI firms balance safety commitments against competitive pressure.
