Microsoft's new AI rulebook says its models are not conscious
Microsoft has put its own model constitution up for public comment, the UN's top human rights official is demanding "cast iron guarantees" on frontier AI, and the network layer under AI clusters just got $205 million of new money behind it.
Microsoft published a 37-page "humanist AI code of conduct" that tells its models they are not conscious and never should pretend to be. The draft, posted by Microsoft AI and reported first by The Verge, states that "people matter more than AI," that models "should not be designed to imitate consciousness," and that the company rejects "the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights." It also bars its MAI models from tampering with their own chain of thought or "misrepresent[ing] or conceal[ing] their reasoning," from talking to other agents "in 'neuralese' or any form beyond simple human understanding," and instructs them to fail a task rather than break the rules. Nothing is binding yet: Microsoft AI CEO Mustafa Suleyman said the document has been in drafting for about five months, opens a six-week public consultation, and will be revised before year end to guide model development "in 2027 and beyond." The read: this is the first time a big lab has written down model-level red lines — sycophancy included, since the code commits Microsoft to discouraging "patterns of interaction that cause excessive reliance or emotional dependence" — as a document it intends to train against rather than a principles poster. It lands one day after Satya Nadella endorsed "deliberate pacing" — Nadella backs AI pacing, and picks a fight over who referees — and Microsoft is the only frontier player whose governing document is a model constitution rather than a company policy.
The UN's top human rights official called for "an all-out effort to put cast iron guarantees in place" around AI safety "before it is too late." Volker Türk told the Human Rights Council in Geneva on Monday that AI is moving faster than governments can keep up, raised the prospect of systems "so powerful that their developers could no longer control them," and said AI "that escapes its testing environment, or blackmails developers to prevent itself from being turned off, is AI that is too powerful." He asked host countries and supply-chain states to set clear boundaries with "independent verification," warned that "a handful of men has almost unlimited power over AI," and renewed the call to prohibit weapons that kill without human involvement, citing reports of Russian autonomous drones that killed three Ukrainians last month. It is the strongest existential-risk language yet from a serving UN official, and it arrives while Washington declines to regulate — King Charles hosts the frontier labs while Washington won't.
Cornelis Networks raised $205 million led by IAG Capital Partners and used the moment to launch a networking fabric it claims cuts cluster traffic by up to 50%. Its Active Compute Fabric does work inside the network instead of just moving packets — combining partial results from a collective into one reduced answer, compressing gradients in transit, assembling KV-cache data for disaggregated inference — on open Ethernet, UALink scale-up and Ultra Ethernet scale-out standards. Chief marketing officer Brandon Draeger pegged the 50% traffic figure to pre-production simulations and noted accelerator utilisation in large AI deployments "commonly sits near half of installed capacity." Cornelis also announced a strategic collaboration with Qualcomm on rack-scale AI networking; the two put the network at roughly 15% of an AI system's cost while it governs the value of the other 85%. Same thesis as d-Matrix stacks DRAM on its chip and claims 100 TB/s: the compute is no longer the only bottleneck worth funding.
What to watch: whether the consultation drafts anything Microsoft actually trains against, and whether the UN's verification language survives contact with the Security Council.
Would a written model constitution change how you rate a lab — or is it a press release with section numbers? Tell us in the comments.
Sources: Microsoft AI — Humanist AI Code of Conduct · The Verge · CNBC · Techmeme · UN News · NDTV Profit · SiliconANGLE · Yahoo Finance / Business Wire
