A new Hugging Face essay argues that Constitutional AI gives private AI developers too much power to set the principles their systems follow. Its central question is less whether those principles sound agreeable than who gets to choose, challenge and amend them.
Anthropic Watch analysis
What happened
In a community article published on 4 October, Vasileios Stergiou compares Anthropic’s Constitutional AI approach with political ideas from Plato and Aristotle. He argues that a private company’s written principles, used to shape model behaviour, are not the same as a constitution agreed through public procedures.
The essay calls for institutions using AI to be able to inspect the principles behind a model, choose their own grounding material and know how those principles can be amended. Read Stergiou’s essay on Constitutional AI.
Why it matters
Model behaviour reflects choices about what counts as helpful, harmful or acceptable. Those choices matter particularly when universities and public bodies use AI in teaching, research or policy work. Stergiou’s essay presses a useful governance question: convenience in adopting a model does not settle who should have authority over its values.
Our read
This is an argument, not a neutral audit of Anthropic’s processes. But it points to a real distinction worth keeping in view: publishing principles is not the same as giving users a say in how they are set or changed. A changelog can tell you what moved; it cannot, by itself, tell you who had the right to move it.
What to watch
- Whether AI providers publish clearer records of changes to the principles shaping model behaviour.
- Whether universities and public bodies demand ways to inspect or configure those principles.
- How providers respond to calls for more public input and meaningful amendment procedures.
Discussion spark: When an institution adopts an AI model, should it be able to set the values guiding that model, or is the provider’s own constitution enough?
Sources and evidence
- Constitutional AI Is Just Plato's Republic With Better Marketing (4 October 2026, 12:12 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.