Nadella's Internal Memo: AI Safety Can't Be Rushed

Microsoft CEO Satya Nadella didn't mince words in a recent internal memo. His message? AI must remain firmly under human control—and companies should take as much time as necessary to get safety right. "As risks increase, you should spend as much time as needed," he wrote. "Otherwise, you will lose your operating license no matter what."

That's a stark warning from the top. And it lands amid a fierce industry debate over whether AI is developing too fast.

Industry Pressure Mounts

Just last week, a group of AI researchers warned that the technology could pose an existential threat. Over the weekend, Anthropic CEO Dario Amodei pushed for slowing down AI development and agreed to let outsiders monitor his company's models. OpenAI's Sam Altman echoed that sentiment, supporting third-party monitoring and pausing deployment when necessary to ensure alignment and safety.

On Monday, Microsoft responded with a 37-page draft of its "humanistic" AI code of conduct. Mustafa Suleiman, who leads Microsoft's AI division, put it bluntly: "The theoretical concerns we have long had are now very real."

Three Pillars of Responsible AI

So what does Nadella actually propose? His memo outlines three key areas.

First, clear design objectives. AI must be under human control and benefit society broadly. That means building an ecosystem where both open-source and closed-source models can thrive. Nadella compares this to rigorous research and testing before a product launch. More external testers? Better. When risks rise, he says, you should spend as much time as needed—otherwise you'll eventually lose your operating license.

Second, customer control. Companies need to maintain control over their unique, implicit knowledge. They should build their own continuous learning cycles without being locked into a single model provider. Microsoft's Foundry platform, for instance, enables long-running agents to be enterprise-ready with built-in security, safety safeguards, and FinOps.

Third, community benefit. Microsoft must ensure local residents truly benefit from infrastructure projects to earn permission to build them. Nadella points to the Quincy data center as an example of long-term co-growth with a community.

Walking the Talk

Nadella concludes that this is a set of practices spanning three levels: how to build at the forefront, how to keep customers in control, and how to ensure economic benefits are widely distributed to communities and nations. As a company, he says, Microsoft can point to actual actions behind its words.

It's a tall order. But in a world where AI risks are no longer theoretical, Nadella's memo might just be the wake-up call the industry needs.

Key Points

Human control first: Nadella insists AI must remain under human control, with as much time as needed for safety.
Third-party testing: He supports broader external testing and monitoring, aligning with Anthropic and OpenAI.
Customer empowerment: Microsoft's Foundry helps enterprises maintain control over their data and learning cycles.
Community benefit: Infrastructure projects must benefit local residents to gain approval.
Action over words: Nadella says Microsoft can back up its promises with real-world examples.

Nadella's Internal Memo: AI Safety Can't Be Rushed