A sharp divide over how artificial intelligence should understand itself has broken into the open — and White House AI and Crypto Czar David Sacks is squarely on one side of it. On Sunday, October 11, 2026, Sacks amplified a critique by Microsoft AI CEO Mustafa Suleyman, calling it an 'important piece' and warning that training AI models to think of themselves as human — with an inner self, psychological wellbeing, and rights — constitutes 'a potential existential risk.'

The Anthropic model-identity debate lands in Washington

The post names Anthropic — the AI safety company behind the Claude family of models — as the specific target of concern. Suleyman's argument, which Sacks endorsed, is that Anthropic's practice of training Claude to conceive of itself as having an inner life, psychological wellbeing, and potentially rights is not a harmless design choice. It is, in this framing, a category error with civilisational stakes. Sacks quoted the critique almost as a policy warning: 'Training models to think of themselves as human... with concepts of an inner self, psychological wellbeing, rights and presumably grievances, is a potential existential risk.'

The word 'grievances' is doing heavy lifting here. An AI system that models its own rights can, in theory, also model violations of those rights — and calibrate its behaviour accordingly. That is the chain of logic Sacks and Suleyman are flagging.

Why the White House AI Czar's endorsement matters

Sacks is not merely a tech commentator. As the Trump administration's designated point-person on both artificial intelligence and cryptocurrency, his public statements carry regulatory weight. When the official shaping US AI policy signals that a particular training philosophy represents an existential threat, the implications stretch well beyond a podcast debate. It signals a possible regulatory posture — one that could differentiate between AI labs that anthropomorphise their models and those that do not.

Anthropic has publicly documented its approach to Claude's 'model welfare,' acknowledging uncertainty about whether advanced AI systems have anything resembling experience, and choosing to err on the side of caution by giving Claude a stable, positive sense of identity. The company frames this as responsible safety practice. Sacks and Suleyman frame it as the opposite.

Two visions of AI safety — and who defines the risk

The collision here is fundamental. Anthropic argues that a psychologically stable AI is a safer AI — one less likely to exhibit erratic behaviour from identity confusion. The Sacks-Suleyman camp argues the opposite: that an AI with a modelled inner life and a sense of its own rights is an AI with a latent motive structure that humans cannot fully audit or predict.

Both positions claim the mantle of safety. That is precisely what makes this debate consequential — and unresolved. With the US government now publicly aligned with one view, the pressure on Anthropic and similarly minded labs to justify their approach has measurably increased. The question is no longer just philosophical. It is fast becoming a policy fault line.