The UAE's Groundbreaking Move: Deploying Agentic AI for Competition Enforcement

A New Era in Regulatory Frameworks

The UAE is deploying agentic AI for competition enforcement and merger reviews. This initiative is set to redefine how governments regulate markets and ensure they remain competitive. The UAE has taken a bold step that no other competition authority has attempted. Shortly after announcing that half of all regulatory work will be managed by AI within two years, the Ministry of Economy and Tourism issued a call for procurement. This isn't a pilot or sandbox, but a genuine mandate to create an AI platform capable of handling merger control assessments, preliminary antitrust investigations, and market studies at a national level.

This approach signifies a structural shift in how a state exercises one of its most critical powers: determining market competitiveness.

The Efficiency Argument

The efficiency argument for AI in regulation is compelling. Antitrust investigations are notoriously slow, resource-intensive, and often lag behind the innovative markets they aim to govern. An AI platform that can compress analytical timelines will free human regulators to focus on judgment rather than process. This also allows regulators to become AI-native themselves.

The UAE has a proven track record of transforming policy ambitions into operational realities faster than many jurisdictions. But, for this operationalization to succeed and serve as a replicable model for regulatory roles assigned to AI globally, three essential guardrails must be established.

The Timeline Paradox

In the realm of regulation, speed is often equated with efficiency. In merger control, time is money. However, regulatory speed should not compromise legal rights. Statutory rights for merging parties, such as access to files and the opportunity to respond to objections, do not automatically diminish just because the regulator's analytical capacity is faster.

If an AI tool processes millions of data points and concludes that a merger is anti-competitive, parties still need a human-readable window to challenge that logic. This creates a timeline paradox: the regulator's clock may accelerate, but the parties' statutory rights remain unchanged. Speed is a virtue, but due process is the law. Accelerated processing should empower parties to move faster, not restrict their rights.

The Verification Trap

The RFI envisions an AI system that will "receive, process, analyse, and generate case outputs ready for human verification". This balances efficiency and responsibility, but it also introduces a crucial risk. AI excels at cognitive tasks, working through checklists and applying established frameworks to structured data. However, the most vital antitrust interventions often require creative thinking.

Identifying why an acquirer paid billions for a deal before the target had commercialized its product is a complex task. The cases involving digital platforms are not straightforward applications of precedent; they require creative interpretation of the law. If machine-led cognition becomes the norm, creativity may become optional.

When regulators are presented with AI-generated outputs that are highly accurate, they may feel compelled to verify rather than probe. This raises the question: who will identify harm that lacks precedent? Governance must ensure that human oversight does not become a mere rubber stamp. Regulators must retain the capacity to envision how markets might evolve, rather than only assessing how they functioned in the past.

The Battle of the Bots

The verification trap poses significant risks, especially in an adversarial context. The moment creativity atrophies is precisely when external actors will exploit it. The deployment of AI in regulatory supervision will inevitably trigger an algorithmic race. If regulators use AI to identify harm, sophisticated legal practices will deploy AI to avoid it. Law firms are already equipped with advanced tools to enhance their efficiency.

We are entering an era where generative adversarial networks will play out not just in research labs but also in regulatory and court filings. Defence teams will use their models to stress-test submissions before filing, optimizing market definitions and competitive narratives until their AI predicts a high probability of clearance. Filings will be designed to pass AI filters.

This is where the “Human In, On, and Over the Loop” framework becomes essential. The human in the loop catches errors in individual cases. The human on the loop monitors whether the system is being gamed across cases. The human over the loop retains the authority to override clearance recommendations when novel theories of harm have not yet been encoded in the model. Each layer is necessary; none is sufficient alone.

The Future of AI in Regulation

With this RFI, the UAE is not just building a tool; it is writing the first chapter of what an AI-enabled regulator looks like at a national scale. The guardrails it establishes will inform policymaking beyond antitrust, shaping the architecture of AI governance globally.

The question is not whether AI belongs in the regulatory state - it most certainly does. The real question is how the regulatory state should govern itself and prepare for what comes next.