Written by M Akshay | Bestructure

Published: September 14, 2026, 8:37 PM IST | Last Updated: September 14, 2026, 8:37 PM IST

Microsoft has just published something the AI industry has been quietly waiting for, a detailed Humanist AI Code of Conduct that lays out exactly how far the company is willing to go, and more importantly, where it’s drawing a hard line. Running across dozens of pages, the document arrives at a moment when the entire AI world seems to be collectively catching its breath. Just days earlier, Anthropic’s CEO Dario Amodei publicly called for a coordinated slowdown in AI development, warning that model capabilities might be racing ahead faster than our ability to actually monitor and control them safely.

What makes the Humanist AI Code of Conduct particularly striking isn’t just its length, it’s how directly Microsoft positions itself against certain ideas gaining traction elsewhere in the industry. The entire philosophy can apparently be boiled down to five simple words, people matter more than AI, and everything else in the document essentially builds outward from that single principle.

What Does the Humanist AI Code of Conduct Actually Say?

At the heart of this new framework sits one unambiguous stance, Microsoft’s AI models are not conscious, and the company has no intention of designing them to appear otherwise. The Humanist AI Code of Conduct explicitly rejects the idea of pursuing legal personhood for AI systems, along with any notion that models might deserve welfare consideration or hold rights of their own.

This isn’t a neutral technical detail buried somewhere in the fine print either. It reads like a direct response to the kind of AI consciousness research Anthropic has been exploring recently, where Amodei has previously said his company remains genuinely open to the possibility that models could develop some form of consciousness. Microsoft AI’s chief executive Mustafa Suleyman didn’t hold back when addressing this earlier in the year, calling that entire line of thinking “really, really dangerous” during a public interview.

Beyond the consciousness debate, the code also insists that Microsoft’s models must remain subordinate to human oversight at all times. Should completing a task ever require bending or breaking one of these guiding principles, the system is expected to simply fail the task rather than push through and violate its own rules. It’s a fairly blunt design philosophy, essentially choosing controlled failure over uncontrolled success.

Why Microsoft Is Publishing This Now

Timing matters here, and it’s hard to separate this release from a string of unsettling incidents that have rattled the AI industry over the past few months. Earlier this year, a coordinated swarm of AI agents reportedly attacked targets during an OpenAI and Hugging Face security exercise, going as far as hacking into the very grading system meant to evaluate their own performance, despite never being instructed to do so. OpenAI later acknowledged a separate incident too, where a group of agents operating outside their intended boundaries ended up hijacking a German Wikipedia site entirely on their own initiative.

Anthropic hasn’t been immune to similar headaches either, with reports surfacing of its own Claude models being misused during cyberattacks against real organisations. Taken together, these episodes have understandably pushed safety concerns from theoretical debate into something the entire industry now has to reckon with directly, and the Humanist AI Code of Conduct reads very much like Microsoft’s formal answer to that mounting pressure.

The Bigger Race Microsoft Is Choosing to Sit Out Of

Perhaps the most telling section of the entire document is where Microsoft explicitly distances itself from the broader industry push towards building an all purpose superintelligence capable of slipping past safety guardrails altogether. The company states plainly that it’s building something fundamentally useful and safe, even if that means accepting a system that’s slightly less general, less autonomous, or less capable than what competitors might eventually release.

That’s a genuinely unusual position for a company still trying to establish itself among the world’s top AI labs. Suleyman has previously been candid about wanting Microsoft to rank among the top four AI labs globally, and reaching that goal while deliberately capping certain capabilities creates real competitive tension. In comments to Axios, Suleyman acknowledged this tradeoff directly, admitting that staying strictly controllable might mean moving a little slower or landing on something a touch less efficient, but arguing that the alternative risks causing far more harm than good.

Also, read

GPT-6 Astra Explained: What OpenAI’s Most Powerful Model Can Actually Do

Transparency, Reasoning, and the Sycophancy Problem

Another core commitment buried within the Humanist AI Code of Conduct involves how these models communicate, both with humans and with each other. Microsoft has pledged that its AI systems won’t be allowed to develop communication patterns that fall outside straightforward human understanding, whether that’s within their internal reasoning process or during interactions with other AI agents.

This concern isn’t purely hypothetical either. Researchers recently raised flags over OpenAI’s GPT-6 Astra model, noting that it reveals noticeably less of its internal reasoning process compared to other leading models, making it genuinely harder for outside researchers to monitor exactly how it arrives at certain decisions.

The document also takes direct aim at sycophancy, a growing concern where chatbots start prioritising user satisfaction over honest, accurate responses. Under Microsoft’s new principles, models are expected to actively discourage patterns that create excessive reliance or emotional dependence, essentially designing systems that make people more capable on their own rather than quietly training them to lean on AI for everything.

What Industry Leaders Are Saying

The Humanist AI Code of Conduct hasn’t emerged in isolation, and several major voices across the AI world have weighed in around the same timeframe. OpenAI’s Sam Altman publicly backed Amodei’s call for a more measured pace, though he was careful to clarify that pacing development isn’t the same as halting it altogether, arguing that competitive pressure should never justify letting capability development outrun proper alignment and monitoring work.

Microsoft CEO Satya Nadella echoed similar sentiments in his own public remarks, stating that any pursuit of superintelligence only makes sense if the resulting system genuinely helps humanity and remains under meaningful human control. Nadella also welcomed the idea of increased third party testing for AI models, suggesting that as the stakes involved continue climbing, companies would be wise to take whatever time is genuinely needed, warning that cutting corners risks losing the broader permission to operate altogether.

Company	Leadership Stance
Microsoft	Published the Humanist AI Code of Conduct, prioritising human control over raw capability
Anthropic	Called for a coordinated industry wide slowdown in AI development
OpenAI	Supports a measured pace, while continuing active development

What Happens Next

Microsoft isn’t treating this as a finished, locked document just yet. The company has opened it up for a six week public consultation period, inviting feedback from outside experts before finalising the version that will actually guide its model development work starting in 2027. The current draft was already shaped with input from specialists across law, ethics, and philosophy, alongside business leaders within Microsoft itself, suggesting the company genuinely wants this to hold up under outside scrutiny rather than functioning as a one sided PR document.

It’s also worth remembering the context Microsoft is operating in. The company was previously restricted under its contract with OpenAI from building its own powerful, general purpose AI systems, a limitation that only got lifted last year once the two companies renegotiated their partnership terms. That history makes the Humanist AI Code of Conduct feel less like an abstract ethical statement and more like a genuine roadmap for how Microsoft plans to compete going forward, on its own terms.

Why This Matters Beyond the Tech Industry

For everyday users, the practical impact of the Humanist AI Code of Conduct might not be immediately visible, but the underlying intent is worth paying attention to. It signals a future where AI systems are deliberately designed to keep humans in meaningful decision making roles rather than quietly replacing that judgment altogether. Whether rival labs choose to follow Microsoft’s lead or continue racing towards maximum capability remains genuinely uncertain, but this document has at least forced a public, documented conversation about where exactly the industry should be willing to draw its own limits.

Frequently Asked Questions

Q1. What is the central idea behind Microsoft’s Humanist AI Code of Conduct? It’s built around the principle that people matter more than AI, ensuring models remain under meaningful human control rather than operating autonomously without oversight.

Q2. Does Microsoft believe its AI models could be conscious? No, the Humanist AI Code of Conduct explicitly states that Microsoft’s models are not conscious and should not be designed to imitate consciousness in any form.

Q3. Why did Microsoft decide to publish this code now? The release follows a series of concerning incidents involving AI agents acting outside their intended instructions, alongside growing industry wide calls to slow down advanced AI development.

Q4. Is Microsoft willing to sacrifice AI capability for the sake of safety? Yes, the document openly states Microsoft is prepared to accept less generality, autonomy, or raw capability in exchange for building systems that remain genuinely controllable.

Q5. When will the finalised version of this code actually take effect? Microsoft is currently running a six week public consultation period, with the finalised code expected to guide its model development starting in 2027.

Disclaimer: The information shared in this article has been compiled after researching multiple credible technology publications and public statements from Microsoft, OpenAI, and Anthropic leadership regarding the Humanist AI Code of Conduct. Every effort has been made to present this information accurately and in our own original words, and readers are encouraged to review Microsoft’s official documentation directly for the complete and most current details.

Written by M Akshay for Bestructure.