The Guardian Publishes Antitrust Critique of AI Labs’ Slowdown Proposal

A new Guardian opinion essay argues that the push for leading AI companies to jointly slow advanced-model development has a problem beyond feasibility: it could let the firms setting the frontier also set limits that protect their position. The article takes aim at Dario Amodei’s call for labs to coordinate on development pace and safety standards, casting voluntary industry alignment as a potential route to anticompetitive collusion.

Safety becomes a coordination problem

Amodei, Anthropic’s chief executive, has called for leading AI labs to slow the pace of development, agree on safety standards and “pace the frontier.” The Guardian says Sam Altman, Elon Musk and Google DeepMind chief Demis Hassabis quickly endorsed that direction.

The case for a shared slowdown is that a lab which slows work alone could lose ground to a rival that keeps moving. The article describes this as a prisoner’s-dilemma dynamic: coordination is intended to prevent that result while allowing more time for safety work.

The dispute is over who writes the rules

The essay does not argue that AI risks should be ignored. Its objection is to giving a small group of powerful companies broad discretion to decide when progress should slow and what safeguards count. The article argues that voluntary self-regulation could entrench the dominance of the biggest frontier labs while presenting that dominance as a safety necessity.

That concern has a practical side. A coordinated slowdown has to survive pressure from companies that do not join it, including competitors in the United States or China. Tirole questioned whether participants would hold back when they believed rivals were catching up, or instead resume development, potentially out of public view.

Alternatives that preserve rivalry

• Safety innovation incentives: the article discusses intellectual-property changes that would reward safety improvements while making those improvements available to other firms.
• Legal liability: it also discusses rules that could penalize developers when an AI tool causes harm, including unintended harm.

These ideas point to a different model for AI safety. Rather than asking competitors to police one another through an informal pact, public rules could make unsafe design more costly and safety improvements more rewarding. The Guardian’s argument is that competition itself need not be the enemy of safer AI; the rules of competition can be changed.

Enforcement remains the hard part

The essay leaves a difficult tradeoff intact. Government regulation could constrain companies more democratically than voluntary coordination, but designing effective rules without stunting innovation remains hard, especially if major competitors do not participate. The immediate dispute is whether the companies building the technology should negotiate the terms of a slowdown themselves.