The news: Microsoft CEO Satya Nadella recently took aim at Anthropic’s Claude Fable 5 AI model, criticizing its high rate of content refusals as overly “editorially controlled” and counterproductive for a creation tool, per CNBC. 

• Nadella was responding to a catch that researchers discovered within Fable 5. The model quietly downgrades its response quality whenever it detects users working on sensitive subjects like advanced AI research or cybersecurity, per TechCrunch. 
• Fable 5, introduced in June, initially aimed to reduce false positives but then added US export controls and new safeguards that “flag a slightly higher fraction of harmless requests,” per CNBC.

Zooming in: Nadella’s take is that enterprises need cost-efficient, custom models that keep internal data private, rather than rental dependencies. Basically, businesses should build their own AI models.

There’s also a competitive angle—Microsoft sells its own coding model and operates Foundry with 11,000 alternatives. The company has also been sunsetting many of Copilot’s free features like Deep Research in favor of similar functionality only available in paid tiers, which indicates pressure to monetize AI sooner rather than later.

Nadella’s remarks, made to Microsoft engineers, expose fresh friction with a key partner: Microsoft has invested $5 billion into Anthropic and the AI startup committed to spending $30 billion on Azure. His critique delivers a keen assessment of Fable’s safeguards, yet it conveniently steers users to Microsoft’s Foundry over Anthropic—revealing the deepening entanglement of competing AI giants.

Recommendations for brands: Advertisers and marketers should treat AI platforms more like interchangeable utilities than locked-in ecosystems, building the ability to swap models as costs and performance shift without affecting process or performance.

Nadella implies that as open-source and niche models proliferate, the premium for any single “one‑stop” AI stack will erode, pushing negotiating leverage toward buyers who keep their options open and toolkits flexible.

• Audit AI vendors for refusal rates on creative or commercial prompts to make sure the AI model is performing as promised and not running downgraded versions because its guardrails were triggered.
• Consider Foundry-style marketplaces, such as Amazon Bedrock and Google Gemini Model Garden, to switch models quickly and negotiate rates.

This content is part of EMARKETER’s subscription Briefings, where we pair daily updates with data and analysis from forecasts and research reports. Our Briefings prepare you to start your day informed, to provide critical insights in an important meeting, and to understand the context of what’s happening in your industry. Non-clients can click here to get a demo of our full platform and coverage.