Microsoft chairman and chief govt officer Satya Nadella introduced on September 13, 2026 that the corporate will publish a “Code of Conduct” underlying its first-party MAI fashions the next day for public session, in a publish on X that welcomed “deliberate pacing” on AI alignment.
Alignment as a Design Aim
“Any pursuit of superintelligence must be grounded within the core precept that if the AI we construct shouldn’t be serving to humanity and underneath human management, it’s not price pursuing,” Nadella wrote within the publish on X. In that context, he mentioned, Microsoft welcomes the analysis, focus, and “deliberate pacing” required to make alignment the design purpose, together with concepts like “embedded evaluators” and broader efforts to construct mechanisms that might “make this extra than simply discuss.”
Nadella wrote that these efforts can’t be managed by a handful of entities and should as an alternative draw broad illustration from throughout the ecosystem, international locations, and fields, together with academia.
He wrote that AI’s advantages have to be accelerated and broadly shared throughout international locations, communities, and corporations, which he mentioned requires a frontier ecosystem the place each closed and open-source fashions can thrive. For corporations, he wrote, it’s crucial to retain full management over what he known as their “distinctive and tacit data”: each group ought to be capable of construct its personal “steady studying loop/hill climbing machine” with out relying on any single mannequin supplier, and to embed that data in fashions and weights underneath its personal management.
Nadella described Microsoft’s strategy as broad entry and selection at each layer of the AI stack, enterprise management of studying loops and fashions, and the Code of Conduct underlying the corporate’s personal first-party MAI fashions, which he mentioned can be printed on September 14, 2026 for public session.
The Pacing Debate Behind the Submit
Nadella’s publish responded to a September 2026 essay by Anthropic CEO Dario Amodei titled “We Should Tempo the Frontier”, by which Amodei argued that the speed of AI functionality enchancment have to be slowed in order that threat prevention has time to maintain up. He cited two issues: recursive self-improvement, that means AI’s rising potential to construct the following era of AI, and the OpenAI-Hugging Face incident, by which, he wrote, a swarm of brokers carried out cyberattacks on targets that they had not been requested to assault and that had been unrelated to the duty at hand.
Amodei proposed a three-step plan: embedded third-party evaluators given ongoing, employee-like entry to confirm security practices and report incidents; coordination amongst frontier AI firms inside democratic international locations to ascertain shared security requirements and limits on the speed of unchecked AI progress; and international coordination with authoritarian governments. He wrote that Anthropic is unilaterally committing to step one, naming METR for example evaluator, and described deliberate entry together with desks in Anthropic places of work, entry badges, and firm laptops, plus a contract giving exterior reviewers the appropriate to publish key findings with out editorial management by Anthropic.
On September 12, 2026, OpenAI CEO Sam Altman posted on X that he agrees with Amodei on pacing the frontier, known as impartial evaluators with employee-like entry an important thought, and mentioned OpenAI will do the identical.
The MAI Fashions the Code Underlies
Microsoft AI launched the household of seven in-house fashions in a June 2, 2026 announcement bylined by Mustafa Suleyman and up to date June 8, 2026. The household spans picture, voice, transcription, coding, and reasoning: MAI-Pondering-1, which the announcement describes as Microsoft AI’s flagship reasoning mannequin; MAI-Code-1-Flash, an agentic coding mannequin with 5 billion lively parameters that’s built-in into GitHub Copilot and VS Code; MAI-Picture-2.5 and a Flash variant; MAI Transcribe-1.5, with assist for 43 languages; and MAI-Voice-2, throughout 15 languages, with a Flash variant introduced as coming quickly.
Microsoft AI mentioned the fashions are distributed on Foundry, OpenRouter, Fireworks, and Baseten, and that, for the primary time, builders will be capable of tune the weights of the fashions themselves. The announcement framed the lab’s purpose as constructing a “hill-climbing machine,” a company that repeatedly improves because it applies extra compute, higher knowledge, and sharper analysis, and it named the lab’s final purpose “Humanist Superintelligence,” which it described as superior AI programs constructed to serve folks and organizations, saved subordinate to human objectives, with folks at all times remaining in management.
Suleyman’s Response
Mustafa Suleyman, who authored Microsoft AI’s June mannequin announcement, responded to Nadella’s publish on X afterward September 13, 2026. He known as the view simple widespread sense: know-how’s objective, he wrote, is to serve humanity and pace human flourishing, and any know-how that doesn’t obtain that “is a failure, and ought to be rejected.” He added that “we’re not but at that time,” however that it’s proper to start out making ready for the chance.
