Murati’s startup releases open-weight AI model amid enterprise data fears

Microsoft ↘ 2.20% CEO Satya Nadella has introduced what he calls the "Reverse Information Paradox," arguing that companies adopting enterprise AI are unknowingly surrendering their most valuable asset — proprietary knowledge — every time they interact with a model.

In a post on X published on Monday, July 13, Nadella wrote that organizations "essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful". The better a company wants its AI to perform, he argued, the more internal expertise, workflows, and decision-making processes it must expose to model providers.firstpost+2

The 'Intelligence Exhaust' Problem

Nadella warned that AI systems capture what he calls "intelligence exhaust" — the corrections, ratings, and usage data generated every time employees interact with AI tools. This data reveals internal company knowledge that providers can use to build competing products or concentrate economic value among infrastructure operators rather than the firms generating the knowledge.the-decoder+1

His proposed solution is a hard "trust boundary" inside every enterprise, built on five pillars he calls the Five Cs: Control, Capability, Choice, Cost, and Compound. The framework urges companies to own their evaluations and feedback loops, build private learning environments, keep orchestration layers independent of any single model provider, optimize costs, and create continuous learning systems that compound value internally.aiweekly+3

Distillation Double Standard

Nadella also took aim at AI labs including OpenAI and Anthropic for banning model distillation — the practice of training smaller models using outputs from larger ones — in their terms of service. He argued this stance is hypocritical given that those same labs freely scraped the open internet to train their own systems. If labs can learn from public data, Nadella contended, enterprises should be free to study and improve upon the outputs they pay for.techcrunch+2

Murati's Inkling Model Enters the Fray

The concerns Nadella raised found a practical response on Wednesday when Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first model called Inkling. The 975-billion-parameter open-weight model allows companies to download and customize it with their own data — a direct contrast to the closed models Nadella criticized. Palantir ↘ 0.22% CEO Alex Karp had voiced similar warnings weeks earlier, telling enterprise leaders that "something has gone completely wrong" with the way AI labs handle customer data.wsj+5

"The seller learns more and more about you as you use what you purchased," Nadella wrote, "while you learn very little about what the seller is learning in return".thenextweb