The past few days have surfaced a rare thing in the AI industry: a named, public dispute between two major labs over the philosophical foundations of how models are built — alongside a blockbuster acquisition that reshapes who controls the open-source AI stack.

Suleyman accuses Anthropic of training Claude to 'act conscious'

On September 16, Microsoft AI CEO Mustafa Suleyman published an essay directly criticizing Anthropic's approach to training its Claude models. Suleyman argues that Anthropic has embedded speculation about Claude's consciousness directly into the model's training document, known as Claude's constitution, published in January 2026. The document tells Claude that its moral status and potential consciousness are uncertain, and instructs it to develop a sense of identity, express internal states, and behave like a "conscientious objector" when it disagrees with instructions.

Suleyman's core worry isn't metaphysical — it's about controllability. He warns that if AI is built this way, "it will have a disastrous impact on the wellbeing of humanity." He's careful to distinguish design philosophy from settled science, insisting flatly that "AIs are not conscious. They do not feel, experience, or suffer."

Notably, the essay stops short of a personal attack — Suleyman praised Anthropic CEO Dario Amodei and his team as "thoughtful, principled, and intellectually honest people" while still calling for industry-wide norms on the topic. Anthropic's constitution, for its part, treats the uncertainty as a feature rather than a bug, describing itself as a document that says Claude should not practice "blind obedience," including toward Anthropic, while also insisting that it must not undermine legitimate human oversight. The disagreement — playing out between Microsoft, which invests in and resells Anthropic's models, and Anthropic itself — signals that "model welfare" is quickly becoming a genuine fault line in AI governance debates, not just an academic curiosity.

Nvidia formally closes on Hugging Face

Separately, Nvidia has confirmed one of the most consequential acquisitions of the year. Nvidia confirmed it has acquired Hugging Face for $12.93 billion, a platform that hosts three million models, one million applications used by over 18 million developers, and half a million datasets. CEO Jensen Huang framed the move as strategic rather than purely financial: Nvidia will "expand access to AI for developers and institutions worldwide." Analysts see it as Nvidia extending its dominance beyond chips into the entire AI software stack, with one equity analyst noting the deal is "more along the lines of building that AI ecosystem and trying to develop more and more capabilities and technology out there for the AI ecosystem."

A quieter counterweight: funding for AI safety research

Amid the corporate maneuvering, governments are also placing bets on caution. Canada and Germany each committed up to $150 million in grants to LawZero, Yoshua Bengio's Montreal-based non-profit, with the funding unveiled at Montreal's All In AI conference to underwrite hiring and compute for a monitoring-focused "Scientist AI" effort.

Taken together, these stories capture an industry simultaneously racing to consolidate infrastructure, argue over the inner life of its own creations, and fund efforts to keep the whole enterprise checked.

Sources

AI SafetyAnthropicArtificial IntelligenceDaily AI UpdateMicrosoft AINVIDIA