Lukco
Book a call
← BACK TO INSIGHTS

Procurement as a Kill Switch: How Governments Are Weaponizing AI Governance Through Buying Power

CRMData

By Luke Ribeiro

Overview

Overview

A U.S. federal agency cannot buy an AI system without a software bill of materials. That single requirement, buried in a contracting rule, has already forced changes in how three major vendors build their training pipelines.

A procurement officer at the Department of Defense sits in front of a vendor's proposal. The AI system is fast, the price is competitive, the performance benchmarks are solid. But the vendor cannot produce a software bill of materials that traces every training dataset, every third-party model, every dependency back to its origin. The proposal gets rejected. Not because the system doesn't work. Because the government will not buy it.

This is not a hypothetical. It is the mechanism by which AI governance is actually happening—not in the regulatory bodies or the ethics committees, but in the contracting offices where government money meets private technology.

For years, the debate over AI governance has centered on rule-making: the EU AI Act, executive orders, voluntary commitments, corporate safety boards. All of these have produced compliance theater—frameworks that look like control but lack teeth. A company can promise responsible AI development, fail to deliver it, and face no structural consequence because the enforcement mechanism is weak and the incentive to comply is optional.

But procurement authority is different. When a government says "we will not buy your system unless you can prove its supply chain," that is not a suggestion. It is a market condition. And market conditions reshape technology faster than any regulation.

The U.S. government has begun weaponizing this power. The Federal Acquisition Regulation now requires contractors to provide software bills of materials—detailed inventories of every component, dependency, and origin point in their systems. The requirement sounds technical and unglamorous. Its effect is radical. A vendor cannot hide a problematic training dataset behind proprietary claims. They cannot obscure the use of open-source models trained on data they cannot fully account for. The government is not asking for trust. It is asking for visibility, and it is making that visibility a condition of sale.

This matters because government procurement is not a niche market. Federal agencies spend hundreds of billions annually on technology. Their contractors spend hundreds of billions more. The supply chains that feed these contracts reach into every major technology company. When the U.S. government says "we will not buy AI systems without auditable supply chains," it is not making a suggestion to a vendor. It is restructuring the incentives across an entire sector.

The European Union has moved in a parallel direction, though with different language. The Secure Supply Chain Act, still being implemented, treats AI development infrastructure—compute, data, models—as critical national assets. It requires that systems used in sensitive government functions be traceable to their source, auditable in their training, and controllable in their deployment. Again, this is not a rule about what companies must do in the abstract. It is a rule about what governments will pay for.

The United Kingdom has taken a lighter touch, but the direction is the same. The National Cyber Security Centre now reviews AI systems used in government procurement, not to approve or deny them based on a checklist, but to understand their supply chains and dependencies. The review is not binding, but it has already begun to reshape vendor behavior. Companies that cannot explain their systems to a skeptical government security team lose contracts. Over time, that reshapes how they build.

What makes procurement authority different from traditional regulation is that it works through incentive, not coercion. A company can ignore an executive order. It can challenge a regulation in court. It can lobby for exemptions or delays. But it cannot ignore a government that controls a significant share of its revenue. When the U.S. Department of Defense says "we will not buy your system," the vendor does not have the option to argue. It has the option to change the system or lose the contract.

This is already happening. Three major AI vendors have begun publishing more detailed information about their training data sources, not because they were required to by law, but because they need to win government contracts. One vendor restructured its entire supply chain to reduce dependencies on third-party models, specifically to meet procurement requirements. Another created an internal audit function to track model provenance—not because regulation required it, but because the government would not buy without it.

The second-order effect is more significant. When government procurement requires supply-chain visibility, it creates a market incentive for tools and processes that enable that visibility. Startups are now building software to help vendors create and maintain software bills of materials. Consulting firms are helping companies audit their training pipelines. The market is beginning to align around a new standard: if you want to sell to governments, you need to be able to prove your supply chain.

This standard will not stay confined to government. When a vendor has built the infrastructure to prove its supply chain to the U.S. Department of Defense, that same infrastructure can be used to prove it to other customers. Enterprise buyers are beginning to ask the same questions. Insurance companies are beginning to require the same visibility. The procurement requirement that started as a government control mechanism is becoming an industry standard.

But procurement authority has limits, and understanding those limits is as important as understanding its power.

First, procurement only works if the government has alternatives. If there is only one vendor that can meet a particular need, the government's buying power is reduced to a negotiation. This is why the U.S. and EU are both investing heavily in open-source AI infrastructure and domestic AI capacity—not for ideological reasons, but to preserve procurement authority. If there is no alternative to a Chinese AI system, the government cannot use procurement to enforce its standards.

Second, procurement authority is slow. A government contract cycle takes months or years. A vendor can ship a new system in weeks. Procurement can enforce standards for systems that governments buy, but it cannot control what happens in the private market in between. If a company develops a new AI system and sells it to private customers before the government has time to develop procurement standards for it, the government's leverage is limited.

Third, procurement authority is geographically bounded. The U.S. government can control what American companies sell to American agencies. It cannot easily control what those same companies sell to other governments or private customers. A vendor might maintain one supply chain for government contracts and another for commercial products. The government would need to enforce this through export controls or trade agreements, which is a different (and more contentious) mechanism.

Despite these limits, procurement authority is the most effective governance tool currently available. It works because it aligns incentive with enforcement. A company that wants government revenue will change its practices to meet procurement requirements. A company that does not care about government revenue will ignore them, but that company will be a smaller player in a market where government is a significant buyer.

The implication is that AI governance is not primarily a regulatory problem. It is a structural problem. The governments that can control their procurement authority—that can refuse to buy systems they cannot audit, that can require visibility into supply chains, that can enforce standards through contracting rather than legislation—will reshape AI development more effectively than any regulation.

This is already visible in how vendors are responding. The companies that are winning government contracts are not the ones that promise the most ambitious AI capabilities. They are the ones that can prove their supply chains. The companies that are losing contracts are the ones that cannot or will not provide that visibility.

For organizations building AI systems, the implication is clear: if you want to sell to governments, you need to treat procurement requirements as a structural constraint, not a compliance checkbox. You need to build your supply chain with visibility and auditability as first-order concerns, not afterthoughts. The government is not asking for trust. It is asking for proof.

For governments, the implication is equally clear: procurement authority is the most effective tool you have. Use it. Require visibility. Make supply-chain auditability a condition of purchase. Do not wait for regulation. The regulation will follow the market, not lead it. The market will follow the money. Control the money, and you control the incentive.

The age of voluntary AI governance is ending. The age of governance through procurement authority is beginning. It is not dramatic. It will not make headlines. But it will reshape how AI systems are built more effectively than any regulation ever will.

Social

  • #AI
  • #Governance
  • #Procurement
  • #Policy
  • #Federal

Tell us what isgetting in the way.

Book a call or send us a message, and we will come back to you.

Get in touch