Microsoft's Nadella bows to Trump's language diktat on "Super Intelligence" and uses it to attack OpenAI and Anthropic

Microsoft CEO Satya Nadella has begun using the phrase "Super Intelligence" instead of artificial intelligence, matching terminology favored by President Trump. Trump had previously posted that anyone who continues to say "Artificial Intelligence" should be considered an enemy. The shift appears in Nadella's latest blog post, where he also argues that companies should treat AI models as insider security threats.
On the substance of the post, Nadella writes that firms should regard AI models the way they regard insider risks, not because the models are necessarily malicious, but because any sufficiently capable actor with access to important systems can make mistakes or be compromised, and containment and control architectures must account for that possibility. He claims that today's AI systems offer no way to understand why they produce a given answer or decision, yet those same systems are deployed with access to sensitive data and given the ability to take mission-critical actions on a user's behalf.
His proposed remedy is to separate intelligence from control over that intelligence. He calls for model diversity rather than dependence on a single model, complete logging of all actions, independent auditing, and the ability for humans to shut things down at any time. The most trustworthy system, he argues, is not the one with the best model but the one in which you have to trust the model the least. The idea that models should be treated like insider threats was previously suggested by Google DeepMind, according to the report.
Nadella's cybersecurity framing follows a pattern in his recent writing. He has previously pursued a data-theft angle, accused AI companies of hypocrisy when they crack down on distillation, and said AI models are becoming interchangeable commodities. Microsoft, which is known for a closed ecosystem and aggressive bundling and upselling, has been lobbying for an open AI ecosystem while quietly replacing more capable OpenAI and Anthropic models inside its Office products with cheaper in-house alternatives in order to protect margins. With this latest post, the argument moves from the business practices of rival labs to the technology itself.
The backdrop is a competitive one. The largest AI labs are on the verge of turning their models into the primary computer interface: a user tells the computer what to do, and it does it. Microsoft has long been that default interface, but its own products are tied to legacy software already sold and installed on millions of machines, and the company does not have frontier models of its own. Even if Microsoft positioned itself as a distributor of this new AI interface, it would end up dependent on the AI companies that supply the models, and its margins would take a hit.
A few things are worth keeping distinct here. The terminology change is a public, observable fact: Nadella used "Super Intelligence" where he would previously have written artificial intelligence, and the report ties that choice to Trump's stated preference. The security argument is also directly attributable to Nadella, including the specific proposals around logging, auditing, model diversity and shutdown capability. The interpretation that the post is primarily aimed at eroding trust in OpenAI and Anthropic, and that it serves Microsoft's commercial position, is the report's reading rather than a claim Nadella makes about his own motives. Readers should treat those two layers separately.
It is also worth noting what "Super Intelligence" normally means in the research literature. The term has historically referred to a hypothetical intelligence far beyond human capability, not to today's language models. Applying it to current systems breaks with that established meaning. Whether that matters in practice is a judgment call, but developers comparing vendor messaging should be aware that the phrase is being used here in a marketing and political register rather than a technical one.
The concrete technical asks in the post are familiar from enterprise security practice. Full action logging, independent audit trails, the ability to shut a system down at any time, and avoiding single-model dependence are all standard controls in high-assurance environments, and they map onto patterns already used for privileged access management and insider-risk programs. For teams building on top of hosted models, the practical implication is that these proposals, if adopted, would push toward multi-model architectures, stronger observability around model calls, and documented kill switches. That is a meaningful change from the current default, where many products route most or all of their inference through a single frontier model.
Why it matters: Developers and product teams building on hosted models should watch whether Microsoft's public push for model diversity, exhaustive logging and independent auditing translates into actual platform requirements or procurement terms, since that would affect how AI features are architected. The language shift also signals that terminology around AI is becoming politically contested, which inference suggests may complicate how vendors describe their products in regulated or government-facing contexts. For teams already worried about dependence on a single model provider, Nadella's argument gives a mainstream enterprise voice to concerns that were previously mostly voiced by security practitioners.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Microsoft CEO Satya Nadella now calls artificial intelligence "super intelligence" instead of artificial intelligence, falling in line with Trump's preferred language. In his latest blog post, he frames AI models as insider security threats, taking yet another shot at OpenAI and Anthropic. The real motive is business: Microsoft risks falling behind as AI labs turn their models into the primary computer interface. The article Microsoft's Nadella bows to Trump's language diktat on "Super Intelligence" and uses it to