Who Gets to Define the Rules for AI? Inside the High-Stakes Battle Over Agent Governance and Regulatory Capture

The battle lines defining who governs artificial intelligence have moved decisively from academic white papers into the heart of corporate and antitrust policy. In an unexpected and forceful intervention, Cohere co-founder and Chief Executive Aidan Gomez published a detailed critique examining the closed-door standards initiatives currently championed by market-dominant frontier AI labs. At the center of the dispute is a coordinated push by prominent Silicon Valley developers to secure narrow antitrust waivers, enabling them to jointly establish binding safety thresholds, deployment standards, and collective development paces.

From the vantage point of enterprise builders and sovereign technology providers, this maneuver is less about technological stewardship and far more about commercial entrenchment. Drawing pointed historical analogies to automotive oligopolies and credit rating agencies where safety and validation mandates quietly calcified into regulatory moats, the critique warns that granting an incumbent cartel the power to set the operating rules for autonomous agents directly threatens enterprise autonomy and competitive innovation.

For nearly two years, the AI policy discourse has been dominated by catastrophic risk narratives that emphasize compute-based thresholds and centralized licensing regimes. Under these frameworks, models exceeding arbitrary floating-point operation benchmarks or specialized tool-calling proficiencies are subjected to pre-deployment gatekeeping. While publicly framed as existential protection, the operational reality of these proposals disproportionately burdens open-weights model developers, specialized enterprise platforms, and sovereign national AI initiatives.

By pushing for explicit antitrust exemptions to coordinate safety rules, incumbent frontier developers seek to establish self-administered regulatory bodies. If realized, these private consortia would hold discretionary authority to define what constitutes a safe agent, what tool-calling integrations are legally defensible, and what release schedules are deemed responsible. For enterprises integrating autonomous systems into customer service, proprietary financial workflows, and supply chain optimization, relying on rules designed by consumer-facing frontier providers introduces profound systemic vulnerabilities.

Rather than ceding governance to an alliance of legacy incumbents, enterprise practitioners require a transparent, decentralized model anchored in measurable operational security. A robust alternative framework must rest on four non-negotiable principles:

1. 1.Empirical Risk Scoping: Evaluating models based on demonstrable, real-world failure modes—such as unauthorized data exfiltration, transactional errors, and system prompt injection—rather than unfalsifiable hypothetical scenarios.

1. 1.Architectural Transparency: Requiring verifiable disclosure of evaluation methodology, automated tool capabilities, and safety bounds without demanding the surrender of proprietary training weights or sensitive corporate data.

1. 1.Capability-Scoped Evaluation: Benchmarking models based on the specific actions they are empowered to execute in a production runtime, ensuring that a focused 8-billion parameter code agent is not subjected to regulations intended for multi-modal frontier reasoning clusters.

1. 1.Independent, Conflict-Free Standardization: Entrusting standards to neutral international engineering bodies and transparent open-source consortia rather than self-interested commercial labs requesting legal immunity.

As regulatory scrutiny deepens across North America, Europe, and Asia, enterprise technology leaders cannot remain passive observers. Organizations deploying autonomous workflows must deliberately design for infrastructure independence. This requires deploying modular orchestration layers, adopting open protocol standards such as the Model Context Protocol, and maintaining multi-cloud failover capabilities.

When enterprises retain custody over their model weights, system prompts, and tool boundary permissions, they insulate their operational roadmap from external cartel agreements. The future of autonomous agents must be shaped by open engineering standards, rigorous empirical verification, and verifiable execution—not closed covenants signed behind regulatory curtains.

• Cohere Official Blog: Who Gets to Define the Rules for AI? — Primary Essay by Aidan Gomez
• AI Weekly: Cohere CEO Gomez Calls Rival Labs' Standards Plan a 'Cartel' — Industry Reporting & Policy Breakdown
• Hacker News: Community Technical Discussion on AI Rulemaking and Cartels — Practitioner Debate & Analysis