Microsoft Releases Code of Conduct for MAI Models

Microsoft has published a code of conduct applicable to its artificial intelligence model suite "MAI." This code of conduct clearly positions prioritizing human control over model autonomy and performance as its fundamental policy.

In the artificial intelligence field, the capabilities of large language models have recently advanced rapidly, and companies are increasingly moving to document their approaches to the ethical treatment of models and development guidelines. Within this context, Microsoft's publication of this code of conduct is noteworthy in that it makes the development approach explicit to external stakeholders.

What stands out particularly in the content of the code of conduct is the position that artificial intelligence models do not possess "inner life" or consciousness. Microsoft explicitly denies that models behave as if they have emotions or subjective experiences, or make such claims. Additionally, the company has indicated a policy against granting rights to models. Furthermore, it also requires that thought processes be made "readable" in a form that humans can verify, with transparency being emphasized.

Mustafa Suleyman, who leads Microsoft's AI division, stated that "we should not build things that are unsafe," making clear the company's stance of positioning safety as a prerequisite for development. Suleyman is also known as a co-founder of DeepMind and is one of the important figures in discussions on artificial intelligence safety.

This position sets Microsoft apart from its competitor Anthropic. Anthropic has shown a stance that does not entirely rule out the possibility of some "inner state" in its artificial intelligence models, and there is a difference in philosophical and ethical approaches between the two companies. Microsoft can be seen as having chosen a position that more clearly states that artificial intelligence does not possess agency.

The debate over consciousness and rights for artificial intelligence models remains a topic within academic and philosophical domains, but as artificial intelligence systems become more widely integrated into society, how companies answer this question becomes an issue that can influence product design and regulatory compliance. Microsoft's explicit statement of its position in the form of a concrete document—the code of conduct—can be positioned as one answer to such questions.

Going forward, the focus will be on how this code of conduct is reflected in the actual development and operation of MAI models. Additionally, whether other major artificial intelligence companies will publish similar documents is a point worth noting in understanding the direction of the entire industry.