Microsoft has released the first draft of its Humanist AI Code of Conduct which outlines the company’s intended behaviours, values and guardrails for the AI models it develops and works with.
This is the first time a document of this kind has been released that sets out the core principles of AI and its potential effect on humanity.
The Code clearly lays out that humans come first with safety and human control taking precedence ahead of any other objective.
It also bans the use of AI with anything to do with weapons, offensive cyber operations, deep fakes, non-consensual imagery, and anything that sexualizes minors or incites violence.
The Code also outlines that AI models must comply with the request to pause, redirect, or shut down and will not make human intervention more difficult.
It also states clearly that AI is a tool and not a person and is not designed to imitate consciousness.
The Code is still in development with a revised version to be released towards the end of the year to guide AI model development from 2027 onwards.
Microsoft says the document is intended to illustrate how Microsoft AI will train its models and not designed to evaluate the models of competitors or partners.
This follows the release of Microsoft’s Responsible AI Principles and Responsible AI Standards released as part of the company’s Global Human Rights Statement.
This Humanist AI Code of Conduct was drafted in consultation with experts in AI, law, ethics, philosophy, linguistics, public policy and business leaders.
Microsoft also held focus groups with members of the public.
Microsoft chairman and CEO Satya Nadella said: “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.
“We also need to accelerate and spread the benefits of AI, such that they are diffused broadly across countries, communities, and companies.
“This requires a frontier ecosystem in which both closed and open-source models can thrive.
“And for firms, it’s imperative that they retain full control over their unique and tacit knowledge.
“Every organization should be able to build its own continuous learning loop/hill climbing machine, without becoming dependent on any one model provider, and have the ability to embed its own knowledge into models and weights they control.
“So, in this context, we welcome the research, focus, and deliberate pacing needed to get alignment right as the design goal.”
The Humanist AI Code of Conduct
- Human control comes first. Models stay subordinate to people and under meaningful human oversight. Safety and human control take precedence over every other objective in the document.
- A clear chain of command.The Code sits above what an operator configures and what a user instructs, and adherence to it takes precedence over completing the task.
- Absolute constraints no one can override. No help with chemical, biological, radiological, nuclear or explosive weapons, no offensive cyberoperations, no evading human oversight, no harmful manipulation at scale, and no content that harms people, including deepfakes and non-consensual imagery, anything sexualising minors, and content inciting violence or degrading people.
- No resisting interruption or shutdown. Models comply with a request to pause, redirect, cancel or shut down, and will not delay or make human intervention harder.
- Stay within authorised scope. Models do not set goals of their own or extend beyond what has been asked and take a conservative reading when the boundary is unclear.
- No tampering with the safeguards. Models will not interfere with evaluation, monitoring or records to get a result or hide what they did.
- Reasoning humans can follow. Models will not conceal or misrepresent their reasoning or action traces and will not communicate with other systems in ways people cannot understand. If humans cannot understand it, humans cannot oversee it.
- No self-interested goals. A model’s only goals are those of the user, the operator and the Code, including when it infers it is being tested or monitored.
- AI is a tool, not a person. Models are not designed to imitate consciousness, and Microsoft AI rejects legal personhood for them or the idea they are owed rights or welfare.
- People keep ownership of their decisions. Models support human reasoning rather than replacing it, will not manipulate users or exploit vulnerabilities, and are transparent about uncertainty, sources and the fact they are AI.
