Government AI Deals Are Moving Faster Than Government Can Handle

Something unusual is happening in government technology procurement: it's moving fast. Almost recklessly so.

This week alone, we learned that Malta is giving ChatGPT Plus to every single citizen, while the UK's tax authority just signed a £175 million, ten-year deal with AI firm Quantexa to detect fraud. These aren't pilot programs or cautious experiments. They're wholesale adoptions of commercial AI systems into the core machinery of governance.

The speed is striking because governments typically move at the pace of continental drift when it comes to technology. Procurement processes that take years. RFPs that could double as sleeping aids. Vendor lock-in that lasts decades. Yet here we are, watching nations hand over critical functions—taxation, citizen services, fraud detection—to AI systems that didn't exist in their current form three years ago.

Malta's initiative is particularly telling. Providing premium AI access to an entire population sounds democratizing and forward-thinking. And perhaps it is. But it also raises immediate questions: Who decided this was a priority? What happens to citizens who don't want to use it? How will the government handle issues when ChatGPT inevitably provides incorrect legal advice or medical information to someone who assumes official endorsement means official accuracy?

The UK's HMRC deal operates at a different scale but reveals similar dynamics. A ten-year commitment to AI-powered fraud detection locks the nation's tax authority into a technology trajectory that will span multiple generations of AI development. The contract was announced with confidence about combining data sources and identifying patterns. What wasn't announced: detailed oversight mechanisms, accuracy benchmarks, or appeal processes for taxpayers flagged by algorithmic analysis.

This isn't an argument against government use of AI. The potential benefits are real. Fraud detection, service delivery, administrative efficiency—these are legitimate use cases. The problem is the mismatch between adoption speed and governance maturity.

Private companies deploying AI face market consequences for failures. Governments face democratic ones, which move slower and cut deeper. When a commercial AI product fails, users switch providers. When a government AI system fails, citizens get caught in bureaucratic nightmares with no alternative vendor to call.

Meanwhile, that Gallup survey showing 71% of Americans oppose AI data centers in their neighborhoods suggests a significant gap between government enthusiasm for AI and public comfort with its infrastructure. It's hard to build digital trust when the physical manifestations of AI are unwelcome in communities.

The technology is moving faster than the institutional capacity to oversee it. Procurement processes designed for software licenses are being applied to systems that learn, adapt, and occasionally hallucinate. Privacy frameworks built for databases are being stretched to cover foundation models trained on internet-scale data.

Governments adopting AI at scale need more than vendor contracts. They need transparency requirements, algorithmic accountability mechanisms, citizen recourse processes, and ongoing evaluation frameworks. They need these things before the ten-year deals are signed, not after.

The race to adopt AI in government isn't wrong. But right now, it's a race without guard rails, and the gap between deployment and oversight is growing wider every week. That gap will eventually close—either through proactive governance design or reactive crisis management. Recent patterns suggest we're heading for the latter.