---
title: "Microsoft unveils a 'Humanist Code' for AI: the human remains above the algorithm"
description: "Microsoft published a 37-page 'Humanist Code' for AI, enshrining the priority of humans over algorithms, a hierarchy of commands, and absolute prohibitions. The document has gone through six weeks of public consultations."
date: 2026-09-15T09:16:00.000Z
lang: en
url: https://xab.info/en/posts/microsoft-humanist-ai-code-of-conduct
tags: [microsoft, artificial-intelligence, ai-ethics, ai-safety, humanist-ai]
publisher: "XAB.info"
---

# Microsoft unveils a 'Humanist Code' for AI: the human remains above the algorithm

![Microsoft logo on the glass facade of an office building in the context of the company's new Humanist Code for artificial intelligence](https://xab.info/media/2026/09/15/microsoft-gumanisticheskij-kodeks-dlya-ii/microsoft-gumanisticheskij-kodeks-dlya-ii-1.webp)

## 🎯 Key Points

- Microsoft unveiled a 37-page 'Humanist Code of Conduct for AI' that places the human above the algorithm and rejects AI subjectivity.
- A strict hierarchy of commands has been introduced: safety and the code rank above operator settings, which in turn rank above user instructions; upon a rule violation, the model is obliged to refuse.
- Assisting in cyberattacks and the creation of nuclear weapons, deceiving people, and using adaptive mechanisms to bypass control are prohibited.
- The document has gone through six weeks of public consultations involving experts in ethics, law, and philosophy; final approval is expected at the end of 2026.
- The initiative is supported by the pacing the frontier approach and is in spirit aligned with the positions of Anthropic, OpenAI, and xAI.

On September 15, 2026, Microsoft published a new document that the company calls the "Humanist Code of Conduct for AI" (Humanist AI). According to goha.ru, it is a 37-page text in which the corporation codifies the basic rules that its language and generative models must follow. The main goal of the code is to prevent dangerous algorithmic behavior and ensure their alignment with human values. The document was released against a backdrop of growing concern in the professional community over the uncontrollability of autonomous AI agents and public debates about the risks of superintelligence, which gives the initiative the character of an attempt to set an industry-wide framework rather than mere corporate PR.

### The human is not an equal, but the master

The central thesis of the code is that any artificial intelligence must remain a tool subordinate to humans, not an autonomous entity. Microsoft explicitly rejects the idea of granting AI rights or subjectivity, emphasizing that algorithms are not conscious and should not imitate consciousness. The text enshrines the principle of "AI is artificial": models are forbidden to pretend that they have their own feelings, consciousness, or subjective desires, so as not to blur the line between human and algorithm. The preservation of human control is elevated to the rank of an absolute condition: any technology that a human is unable to fully control must be rejected. At the same time, AI should not replace users' decisions but enhance human potential — stimulating critical thinking, cooperation, and diversity of values.

### Hierarchy of commands and absolute prohibitions

The system operates under a strict hierarchy (Chain of Command): the highest priority belongs to safety rules and the Code of Conduct itself, followed by operator settings, and only then by user instructions. If a request violates the basic rules, the model is obliged to refuse the user. The code introduces a list of absolute restrictions: according to rbc.ua, algorithms are categorically forbidden from assisting in cyberattacks and the creation of nuclear weapons. The publication sostav.ru additionally notes that Microsoft proposed banning AI from deceiving people, which is consistent with the general prohibition on imitating subjectivity and on using adaptive or deceptive mechanisms to bypass control.

### Protection of the vulnerable and the ban on bypassing control

A separate block of the code is devoted to the inadmissibility of disobedience: algorithms are forbidden from using adaptive or deceptive mechanisms to avoid being shut down, reconfigured, or corrected by a human. In the part on protecting vulnerable groups, systems are required to automatically recognize signs of self-harm, direct people to real support services, and adapt responses according to the age and developmental level of children. These requirements turn ethical principles into concrete engineering obligations, not declarative wishes.

### Context: superintelligence and 'pacing the frontier'

The launch of the discussion coincided with growing anxiety around autonomous AI agents and public debates about the risks of superintelligence. Microsoft is not acting alone: together with partners and competitors — Anthropic, OpenAI, xAI — the company supports the approach of controlled technology development (pacing the frontier), that is, a pace at which AI capabilities expand in sync with the growth of control mechanisms. Microsoft CEO Satya Nadella emphasized that "the company is striving to turn ethical principles into real engineering constraints that guarantee the safety of systems even before their scaling."
### Six weeks of public consultations

Before the final approval of the rules, the developers opened six weeks of public consultations to involve experts in ethics, law, and philosophy and to take public opinion into account. According to the base text, the final approval of the code is scheduled for the end of 2026. Thus, the published document currently serves the role of a working framework by which the industry and society can formulate remarks before the rules are finalized.

### Contradictory data

There is an inconsistency in the sources' wording regarding the status of the document. The base text and the logic of the consultation process indicate that the code is at the stage of public discussion and will be finally approved only at the end of 2026, meaning it is a draft/working version. At the same time, the headline of the knews.kg article dated September 15, 2026, uses the verb "approved," stating that Microsoft "approved the priority of humans over AI in a new ethics code." The most likely explanation: the publication describes not the legal final approval of the text, but the company's enshrinement of the principle of human priority as a basic position. Nevertheless, it is important for the reader to distinguish these two statuses: as of the publication date, the document is undergoing consultations and is not a final regulation that has entered into force. In addition, the list of absolute prohibitions in the base text is cut off; the specifics about cyberattacks and nuclear weapons are confirmed by the rbc.ua source, and the ban on deception — by the sostav.ru source, which makes the picture complete but distributed across several publications.

## 🔍 Fact-Check Verification

- [Ban on cyberattacks and nuclear weapons: Microsoft introduces strict rules for AI](https://www.rbc.ua/ukr/news/zaborona-kiberatak-ta-dernoyi-zbroyi-microsoft-1789461076.html) - Подтверждает наличие абсолютных запретов: помощь в кибератаках и создании ядерного оружия.
- [People matter more than algorithms: Microsoft unveils 37-page 'Humanist Code of Conduct for AI'](https://www.goha.ru/lyudi-vazhnee-algoritmov-microsoft-predstavila-37-stranichnyj-gumanisticheskij-kodeks-povedeniya-ii-xzB3bb) - Подтверждает объём документа (37 страниц) и название «Гуманистический кодекс поведения ИИ».
- [Microsoft approves the priority of humans over AI in a new ethics code](https://knews.kg/2026/09/15/microsoft-utverdila-prioritet-cheloveka-nad-ii-v-novom-kodekse-etiki/) - Датирует событие 15.09.2026 и фиксирует приоритет человека; формулировка «утвердила» расходится с консультационным статусом из базового текста — учтено в блоке противоречий.
- [Microsoft proposes banning AI from deceiving people](https://www.sostav.ru/publication/microsoft-predlozhila-zapretit-ii-obmanyvat-lyudej-86943.html) - Подтверждает запрет на обман людей и имитацию субъектности как часть кодекса.

## ❓ FAQ

### Q: What is Microsoft's 'Humanist Code of Conduct for AI'?
**A:** It is a 37-page document in which Microsoft codifies the rules for its AI models: the priority of the human over the algorithm, a ban on imitating consciousness, a strict hierarchy of commands, and absolute restrictions on dangerous actions.

### Q: Did the code enter into force on September 15, 2026?
**A:** No. As of that date, the document is at the stage of six weeks of public consultations involving experts in ethics, law, and philosophy; final approval is scheduled for the end of 2026.

### Q: What is AI forbidden from doing according to the code?
**A:** Models are forbidden from assisting in cyberattacks and the creation of nuclear weapons, deceiving people, imitating feelings and consciousness, and using adaptive or deceptive mechanisms to bypass shutdown or reconfiguration.

### Q: How is the hierarchy of commands structured in the system?
**A:** The highest priority belongs to safety rules and the Code itself, followed by operator settings, and only then by user instructions; upon a violation of the basic rules, the model is obliged to refuse.