Nadella Switches to "Superintelligence" and Calls AI Models an Insider Threat

Satya Nadella no longer says "artificial intelligence." In his latest blog post, the Microsoft CEO uses "superintelligence," and according to The Decoder, he does so in line with Donald Trump's preferences. In a second move, he describes AI models as an insider security threat and again takes aim at OpenAI and Anthropic. The publication argues that business interests lie behind the wording: the labs are turning models into the main interface of the computer, and Microsoft risks losing the role it has built its business on for the past thirty years.

A word as a political signal

A change in terminology looks minor, but big vendors rarely choose words by accident. "Superintelligence" sounds bigger and sits closer to the language set in Washington. For a company whose government contracts and regulatory risks are measured in billions, matching the administration's vocabulary is convenient insurance. The word itself is not new: in late 2025, Microsoft AI, led by Mustafa Suleyman, already announced a team working on "humanist superintelligence." What is new is that the term has moved from a research division into the CEO's personal blog and replaced the basic "AI."

The model as an insider

The second part of the message is more noticeable. Nadella presents AI models as a threat from inside the perimeter: an agent with access to email, files and corporate systems effectively behaves like an employee who can be tricked, bribed with a prompt, or undone by a configuration error. From a security standpoint, the argument is reasonable, and infosec teams have discussed it for a long time.

But the question is who exactly is named as the source of risk. The blow lands on OpenAI and Anthropic, the labs that sell models directly and want to become the user's point of entry. For Microsoft, which sells not only models but also management, identity and access control, this framing is advantageous: the more dangerous the "insider," the more necessary the one who guards it.

The real stake is the interface

The Decoder names the true motive outright: Microsoft may fall behind because the labs are making the model the main interface of the computer. If users talk to an agent instead of opening Word, Excel and Windows windows, value shifts from applications to whoever stands between the person and the task. Microsoft's position here is two-sided:

• it holds a large stake in OpenAI after the 2025 restructuring, but OpenAI increasingly operates as an independent platform;
• it invests in Anthropic and sells its models through Azure, so it earns money from the company it criticizes;
• its core assets, Windows and Office, are exactly what an agent interface could devalue;
• its trump cards in the dispute are enterprise security, compliance and access management.

Hence the tone: Nadella attacks partners without breaking with them. This is positioning, not a split.

How to read such statements

If you are choosing a platform for enterprise agents, separate two things: the vendor's rhetoric and your own security requirements. The insider-risk thesis is worth testing in practice, regardless of who states it. The minimum set looks like this: separate accounts for agents, the principle of least privilege, action logging, restricted access to external channels, and manual confirmation of irreversible operations. All of this applies to models from any provider, Microsoft included.

One last point. The Decoder's article rests largely on interpreting motives, not on documents. Read the full text of Nadella's blog post yourself before drawing conclusions about strategy. But the direction is clear: the fight over who becomes the shell for AI has moved from presentations into public polemics.