Is Government Procurement Fueling an AI Arms Race?

AI ethics and responsible AI are increasingly common topics in public sector conversations. But one category of risk remains largely overlooked: the possibility of catastrophic harms from advanced AI, often referred to as “AI safety.” These include:

• Geopolitical instability due to rival states competing for dominance, sometimes referred to as an “AI Arms Race”,

• Loss of human control over advanced AI system (AI models), and

• Misuse of models by state and nonstate actors, including for terrorism.

These risks are unique because of the scale of harm they could cause and the current lack of active governance and mitigation efforts.

Where AI Shows Up in Public Sector Conversations

1. As a public servant, you likely receive messaging that your job is at risk due to AI either because it will become obsolete, or because more technologically saavy colleagues will overtake you. To support the upskilling that this narrative inspires, Apolitical offers events such as How to Build AI Skills for Your Role, and courses including AI Fundamentals for Public Servants.

2. Public servants are inundated with AI marketing, often encouraged to adopt tools on the sly or promote them for formal, approved use. Business-to-government marketing is alive and well! Today, working in government means evaluating tools introduced through a flood of targeted ads as well as official procurement processes. It’s hard to do your job without wondering what you might be missing if you don’t take a few of them for a spin.

3. In many jurisdictions, public servants are encouraged to learn about AI bias, AI ethics, and, more recently, the climate impacts of AI. Indigenous data sovereignty in the context of AI is also slowly gaining traction in places like British Columbia and New Zealand. The growing awareness of all these issues amongst public servants is incredibly important. But AI safety risks also deserve to be on the menu of concerns that public servants can advocate for without being seen as troublemakers.

So What Can Public Servants Do about AI Safety?

You likely aren’t in a position to research, evaluate, and/or propose policies for regulating AI model development in service of catastrophic risk prevention. You also may work for a jurisdiction that has opted not to pursue AI governance, and is choosing to go full speed ahead with AI adoption. Either way, you're not out of options.

Educate yourself about AI safety, and come to your own conclusions. Take care of your mental wellbeing throughout; the subject matter is heavy. Consider taking this 2-hr free course, AI Safety Fundamentals. If you’re a Canadian federal public servant, the Canada School of Public Service offers Understanding AI Safety in Government. For longer but still introductory reading, check out An Overview of Catastrophic Risks from the Center for AI Safety or Preventing an AI-Related Catastrophe from 80,000 Hours.

Share resources with your team. If you agree that AI safety is under-discussed, share one of the resources above with colleagues, and ask your manager for time to take training about AI safety. Slowly, important resources like AI 2027 are being shared in the AI in Government Apolitical Community of Practice. Browse, and join the conversation when you feel ready.

Ask questions about procurement policies for AI tools. The Ada Lovelace Institute has released a plea for the UK government to establish a National Taskforce for Procurement of AI in Local Government. Policymakers and procurement officers are becoming more aware of the environmental, social, and strategic consequences of their decisions. But what about looking at how procuring for AI tools contributes to reckless model development? Buying from irresponsible AI labs, including labs with no safety commitments, means procurement officers might be contributing to risk without realizing it. If they adopt more cautious standards, this can create demand-side pressure for safety and accountability. Government agencies should be encouraged to evaluate not just the performance and compliance of AI tools, but also the broader consequences of their procurement decisions. Maybe, just maybe, public servants can set the tone for private sector behaviour.

You don’t need to be a machine learning expert to care about risks to humanity. But, you will need to have the courage to ask hard questions and the stomach to handle the unsettling reality of these risks. The conversations are beginning, and you can help to steer them, from your own corner of government.

Views are my own and do not necessarily reflect the position of my clients or employer.