AI Accountability Is Moving Faster Than the Frameworks Around It
I've spent the better part of my career sitting at the place where technology meets public institutions, and if there's one pattern I keep returning to, it's this: the gap between what a tool can do and what society has agreed it should do almost always widens before it narrows. I watched that happen with digital political advertising at Google in the early years. I'm watching it happen again now, but this time with AI, and the stakes are meaningfully higher.
What's different this cycle is the speed. When I was helping shape industry standards on the civic engagement and elections teams at Google, the policy conversations were slow, deliberate, and often reactive. A problem would surface publicly, legislators would hold hearings, and the industry would eventually respond. That rhythm, however imperfect, gave institutions some room to catch up. The AI accountability conversation doesn't have that luxury. Governments are being asked to regulate systems they don't fully control, applied to problems they didn't anticipate, often after those systems are already embedded in public life.
I saw this pressure up close at Dataminr, where I built the AI for Good program from the ground up. We were working with humanitarian organizations and nonprofits applying real-time data and AI to crisis response and early warning systems. The potential was clear. So were the risks. Questions about data sourcing, algorithmic bias, and the ethics of automated threat detection didn't come from regulators first. They came from the civil society partners sitting across the table from us. Those partners were often more sophisticated about accountability than the policy environment around them.
That experience shifted how I think about where accountability actually gets built. It's not in legislation alone. It's in the decisions made at the program design stage, before a product ships, before a partnership is signed. At Dataminr, that meant being deliberate about which use cases we'd pursue and which we wouldn't, and building governance into the program's structure rather than treating it as an afterthought. Those decisions were hard to make and harder to defend internally. But they were the right ones.
What I'm observing now, in 2026, is that more organizations are starting to operate that way, not because they've developed a sudden ethical clarity, but because the reputational and operational cost of getting it wrong has become undeniable. Cities and governments are asking harder questions before signing AI contracts. Nonprofits are pushing back on vendor terms they would have accepted five years ago. This is progress, but it's uneven. The organizations with the capacity and the expertise to ask the right questions are pulling ahead. The ones without that capacity are exposed.
That gap worries me more than the technology itself. At the City of Detroit, where I served as Chief Development Officer, I saw firsthand how resource constraints shape what government can and can't do when it comes to evaluating new tools. Large tech vendors have entire policy teams. A mid-sized city government often has one person trying to cover the same ground. The asymmetry is real. And AI doesn't reduce that asymmetry. If anything, it sharpens it.
So where do I think this is heading? I believe the next two to three years will produce a meaningful split. On one side, you'll have jurisdictions and institutions that have invested in internal capacity: people who understand AI well enough to negotiate with vendors, set conditions, and evaluate outcomes. On the other side, you'll have those that haven't, and they'll find themselves locked into systems and contracts they don't fully understand. The technology will keep moving. The question is whether the public institutions meant to govern it can close that gap before it becomes structural.
That's the problem I want to keep working on. Not just advocating for better policy in the abstract, but helping build the specific human capacity that makes good governance possible. The framework matters. But so does the person in the room who knows enough to push back.