Father Paolo Benanti, an AI adviser to the Catholic Church, is calling for public institutions rather than frontier laboratories to set the rules governing advanced models. He argues that warnings about hypothetical superintelligence can make the technology appear so specialized that only the companies building it are qualified to decide how it should be constrained.
That framing creates a conflict of interest. The same firms asking for controls are competing for investment, customers and technical leadership, and may favor requirements that smaller rivals cannot meet. Benanti compares an industry defining its own shared rules to cartel behavior and says ethical decisions require democratic debate.
His position is not that governments should ignore severe future risks. He favors standards intended to keep systems under human control, but is wary of banning specific development methods in ways that could freeze useful innovation. He also questions whether claims about model capabilities can be separated from the economic and political incentives of the people making them.
The comments respond to proposals such as mandatory third-party evaluations, limits on training models from competitors’ outputs and bans on AI-assisted recursive improvement. Those proposals differ substantially, and no single regulatory design follows from Benanti’s critique. The concrete principle is procedural: laboratories should provide evidence and technical input, but should not have exclusive authority over the obligations that apply to them.