Who gets to define the rules for AI?

TEXT START: Artificial intelligence is remaking the world we live in.

The Dissection

This is a competitor’s regulatory offensive disguised as public-interest governance. Cohere attacks frontier labs for seeking antitrust exemptions and safety-rule control, correctly identifying the mechanism of regulatory capture: incumbents define the threshold, fund the auditors, and convert their existing advantages into legal entry barriers.

But the article’s deeper function is to preserve the legitimacy of AI expansion while contesting who captures it. It wants AI deployed through banks, telecom networks, defense ministries, and infrastructure—provided the rules do not freeze Cohere below the dominant firms. The battlefield is not whether AI reorganizes society. It is which corporate bloc controls the displacement engine and collects the rents.

The Core Fallacy

The article treats better governance as if it can preserve a competitive, broadly beneficial AI market. Under the Discontinuity Thesis, that is the central error.

Evidence-based standards, transparency, independent testing, and international oversight may reduce specific technical risks. They do not solve the structural outcome: AI severs the mass employment → wage → consumption circuit. A more open AI market still concentrates productive power in whoever controls compute, energy, models, distribution, data, capital, and deployment. Competition can change the identity of the Sovereigns; it does not restore the majority’s productive necessity.

The article also assumes that “what an AI system can do” can be cleanly separated from who built it, who operates it, and who controls the surrounding infrastructure. That is false in practice. Capability is not merely a property of a model. It is a property of the model-plus-tools-plus-agents-plus-access-plus-operator stack. The proposed evidence regime therefore cannot escape political judgment or power. Whoever defines contexts, thresholds, observability, incident severity, and acceptable deployment still determines who may scale.

The author is right that “safety” can become cartel armor. He is wrong to imply that procedural pluralism neutralizes the underlying concentration. It merely produces a more polished contest over the same throne.

Hidden Assumptions

• That international coordination can outrun national-security competition. It cannot reliably do so while states treat advanced AI as strategic infrastructure.
• That independent evaluators can remain independent when the firms being evaluated control the technical complexity, access, funding, and career pipeline.
• That published evidence will determine regulation rather than regulation determining which evidence is funded, recognized, and admissible.
• That capability-based rules are neutral. They are not; thresholds and test environments are policy choices with direct effects on market entry.
• That smaller labs, universities, and new entrants can compete if bureaucratic barriers are reduced. Compute, energy, talent, distribution, and security costs still impose massive structural concentration.
• That preventing incumbent capture preserves meaningful competition for society. It may instead create a wider field of firms competing to automate the same human labor.
• That safety is the dominant public question. Under the DT lens, ownership and control of automated productive capacity are at least as decisive as model failure modes.
• That preserving commercial advantage through open competition is socially preferable to preserving commercial advantage through coordinated rules. For displaced workers, both can terminate the same way.

Social Function

Primary classification: transition management and elite self-exoneration, with a substantial partial truth.

The partial truth is real: frontier labs have incentives to inflate catastrophic narratives, define safety in ways that favor their scale, and use regulation to turn private advantage into public law. The cartel analogy is analytically useful for the governance mechanism.

The anesthetic is the implied remedy. By presenting independent review, transparency, proportionate testing, and democratic participation as the route to a safe AI future, the article shifts attention from ownership of automated production to procedure. It asks who gets to write the rules while leaving largely intact the more consequential question: who gets to own the machines that make human labor economically optional?

Cohere’s position is therefore not emancipation. It is altitude selection. The company is attempting to remain a viable Servitor or junior Sovereign by preventing the largest labs from converting safety regulation into a moat. Its anti-cartel language is a weapon in a corporate succession struggle.

The Verdict

This is a sharp attack on regulatory capture wrapped around a strategically incomplete diagnosis. It correctly identifies incumbent safety rules as potential cartel infrastructure, but mistakes fairer access to the AI race for a solution to the social catastrophe produced by the race.

The article argues over who may steer the vehicle while accepting the vehicle’s destination: automated production, concentrated ownership, and the collapse of mass productive participation. Under the Discontinuity Thesis, it is not a survival blueprint. It is a transition memo from one AI firm trying to avoid being buried beneath another firm’s regulatory fortress.