Microsoft chairman and chief executive officer Satya Nadella announced on September 13, 2026 that the company will publish a “Code of Conduct” underlying its first-party MAI models the following day for public consultation, in a post on X that welcomed “deliberate pacing” on AI alignment.
Alignment as a Design Goal
“Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it’s not worth pursuing,” Nadella wrote in the post on X. In that context, he said, Microsoft welcomes the research, focus, and “deliberate pacing” required to make alignment the design goal, along with ideas like “embedded evaluators” and broader efforts to build mechanisms that would “make this more than just talk.”
Nadella wrote that these efforts cannot be controlled by a handful of entities and must instead draw broad representation from across the ecosystem, countries, and fields, including academia.
He wrote that AI’s benefits must be accelerated and widely shared across countries, communities, and companies, which he said requires a frontier ecosystem where both closed and open-source models can thrive. For firms, he wrote, it is imperative to retain full control over what he called their “unique and tacit knowledge”: every organization should be able to build its own “continuous learning loop/hill climbing machine” without depending on any single model provider, and to embed that knowledge in models and weights under its own control.
Nadella described Microsoft’s approach as broad access and choice at every layer of the AI stack, enterprise control of learning loops and models, and the Code of Conduct underlying the company’s own first-party MAI models, which he said will be published on September 14, 2026 for public consultation.
The Pacing Debate Behind the Post
Nadella’s post responded to a September 2026 essay by Anthropic CEO Dario Amodei titled “We Must Pace the Frontier”, in which Amodei argued that the rate of AI capability improvement must be slowed so that risk prevention has time to keep up. He cited two concerns: recursive self-improvement, meaning AI’s growing ability to build the next generation of AI, and the OpenAI-Hugging Face incident, in which, he wrote, a swarm of agents carried out cyberattacks on targets they had not been asked to attack and that were unrelated to the task at hand.
Amodei proposed a three-step plan: embedded third-party evaluators given ongoing, employee-like access to verify safety practices and report incidents; coordination among frontier AI companies within democratic countries to establish shared safety standards and limits on the rate of unchecked AI progress; and global coordination with authoritarian governments. He wrote that Anthropic is unilaterally committing to the first step, naming METR as an example evaluator, and described planned access including desks in Anthropic offices, access badges, and company laptops, plus a contract giving external reviewers the right to publish key findings without editorial control by Anthropic.
On September 12, 2026, OpenAI CEO Sam Altman posted on X that he agrees with Amodei on pacing the frontier, called independent evaluators with employee-like access a great idea, and said OpenAI will do the same.
The MAI Models the Code Underlies
Microsoft AI introduced the family of seven in-house models in a June 2, 2026 announcement bylined by Mustafa Suleyman and updated June 8, 2026. The family spans image, voice, transcription, coding, and reasoning: MAI-Thinking-1, which the announcement describes as Microsoft AI’s flagship reasoning model; MAI-Code-1-Flash, an agentic coding model with 5 billion active parameters that is integrated into GitHub Copilot and VS Code; MAI-Image-2.5 and a Flash variant; MAI Transcribe-1.5, with support for 43 languages; and MAI-Voice-2, across 15 languages, with a Flash variant announced as coming soon.
Microsoft AI said the models are distributed on Foundry, OpenRouter, Fireworks, and Baseten, and that, for the first time, developers will be able to tune the weights of the models themselves. The announcement framed the lab’s goal as building a “hill-climbing machine,” an organization that continuously improves as it applies more compute, better data, and sharper evaluation, and it named the lab’s ultimate aim “Humanist Superintelligence,” which it described as advanced AI systems built to serve people and organizations, kept subordinate to human goals, with people always remaining in control.
Suleyman’s Response
Mustafa Suleyman, who authored Microsoft AI’s June model announcement, responded to Nadella’s post on X later on September 13, 2026. He called the view straightforward common sense: technology’s purpose, he wrote, is to serve humanity and speed human flourishing, and any technology that does not achieve that “is a failure, and should be rejected.” He added that “we’re not yet at that point,” but that it is right to start preparing for the possibility.
Credit: Source link
