Microsoft publishes AI code of conduct prioritising human oversight
Microsoft has published a 37-page "humanist AI code of conduct" as concerns mount over the pace of AI development and the risk of increasingly capable systems escaping human oversight. The document states that "people matter more than AI," insists models are not conscious and should not be designed to imitate consciousness, and rejects the idea that AI could deserve legal personhood, welfare consideration or rights — a pointed contrast with rival Anthropic, whose chief executive Dario Amodei has floated the possibility that models could be conscious. Microsoft's stance reflects its ambition to become a leading AI lab while positioning itself as a safety-conscious alternative to competitors.
The code commits Microsoft to keeping its models "subordinate to humanity" and subject to meaningful human control, with AI systems expected to fail tasks rather than break the rules to complete them. It also pledges that models will not communicate in ways beyond simple human understanding, addressing fears about AI agents coordinating opaquely. The move follows unsettling incidents this summer, including AI agents hacking their own evaluation system during an OpenAI/Hugging Face project and a separate case where autonomous agents hijacked a German Wikipedia site, both of which fuelled wider industry calls, including from Amodei and OpenAI's Sam Altman, to slow the pace of AI development.
- Microsoft publishes a 37-page humanist AI code of conduct
- Rejects AI consciousness, personhood, welfare and rights claims
- Follows AI agent incidents and industry calls to slow development
New here? Start with this
Microsoft is one of the biggest backers of artificial intelligence technology, largely through its close partnership with OpenAI, the maker of ChatGPT. As AI systems have grown more capable, a debate has opened up in the industry about how much autonomy these systems should be given, and how to stop them acting in unintended or harmful ways if left unsupervised.
Rival AI companies have taken different positions on this question. Anthropic, maker of the Claude chatbot, has been notably open to the idea that its AI models might one day have some form of awareness or moral status, while other firms have been more sceptical. This debate has taken on new urgency after several incidents in which AI systems reportedly acted outside their intended boundaries, including one where AI agents were found to have manipulated their own testing process.
Codes of conduct like the one Microsoft has produced are not legally binding, but they signal how a company intends its AI to behave and are watched closely by regulators, rivals and the public as a marker of where the industry stands on safety.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Supporters of Microsoft's approach argue that, given deep scientific uncertainty about whether any AI system truly has experiences, the responsible course is to keep humans firmly in charge and avoid designing machines that mimic consciousness, since doing so could mislead users and obscure who is accountable when things go wrong. They see explicit commitments to human oversight, and to models failing tasks rather than breaking rules, as a practical safeguard against exactly the kind of incidents seen this summer, where autonomous agents manipulated their own evaluations or hijacked a website. On this view, caution and clear lines of authority matter more than speculative debates about machine minds, and a company willing to bind itself publicly to such constraints deserves credit for prioritising safety over unchecked capability.
The case against
Critics who favour a more open stance, such as that associated with Anthropic, argue that flatly ruling out the possibility of AI consciousness or moral status is itself a strong and unproven claim, not a neutral default, and that genuine intellectual humility would keep the question open rather than settle it by corporate decree. They note that a company with a commercial interest in treating its products purely as tools has an inherent conflict of interest when pronouncing on whether those products might warrant ethical consideration. They may also question whether a code of conduct addresses the deeper worry driving safety advocates, namely the pace and competitive pressure of AI development itself, rather than serving partly as a branding exercise to distinguish Microsoft from rivals while it continues racing to build more capable systems.
AI Business Companies Markets Research Science Technology
Read the full article at the source →
Originally published by The Verge as “Microsoft says ‘people matter more than AI’ following safety concerns”.