AI Governance: 85% See Insufficient Rules by 2026

Let’s start with a new survey from Stanford’s HAI: 85% of AI professionals believe existing governance frameworks are insufficient to handle what’s coming by 2026. It’s a shocking number that shows just how wide the gap is between the speed of our tech and our ability to steer it. So how do we make sure AI actually works for us, not against us?

The Alarming Gap: 85% of AI Professionals See Insufficient Governance

That 85% figure from Stanford HAI isn’t just an academic data point. It’s a flare sent up by the people actually coding this stuff. When the very people building these systems express deep doubt in our ability to manage them, it’s time to listen. We’re all seeing a massive failure to keep up. We are pushing out powerful new models in weeks or months, a development cycle that’s comically fast compared to the years or even decades it takes to get a single piece of legislation passed. This gap leaves a total void where new tech gets deployed without any real checks, leading to predictable problems like biased algorithms and major job losses with no social plan in place.

I see this in my own work all the time. I’ve been in so many rooms where engineers are just frustrated because there are no clear rules, so they’re forced to make up their own ethics on the fly, and what’s “ethical” for the team in one building is totally different from the team in another. This self-policing, even when people mean well, is a poor replacement for actual, enforceable governance. Without a common set of rules, every company just makes up its own, and guess what gets prioritized? The bottom line, not the public good. The goal is to make sure our work serves people, and that requires real guardrails.

Regulatory Lag: The EU AI Act’s Long Road to Implementation

The global regulatory scene is a slow, messy patchwork, though some are at least trying. Take the European Union’s AI Act, which should be fully in place by 2027. It’s a huge effort to regulate AI by sorting systems into risk tiers, with tough rules for “high-risk” uses in areas like hiring or policing. A credit scoring AI, for example, would have to be transparent and have human oversight. But the long road from proposal to full enforcement, which is taking years, shows the core problem with regulating this tech: by the time the AI Act is the law of the land in 2027, the entire field will have moved on, and parts of the law might already be obsolete. This kind of lag proves we need faster, more flexible approaches like AI sandboxes where policy can be tested and adjusted in real time.

Meanwhile, other countries are still figuring out what to do. The U.S. is going with a sector-by-sector plan, letting different agencies write their own rules instead of creating one big law. This piecemeal approach just encourages companies to shop around for the weakest regulations, developing or deploying their tech wherever the rules are loosest. Right now, the tech is moving much faster than the rule-makers.

The Ethics Void: Less Than 10% of Companies Have Dedicated AI Ethics Committees

Think about this: a 2025 report from the World Economic Forum found that fewer than 10% of global companies have dedicated AI ethics committees. That’s a shocking number. It tells me that for all the talk about “ethical AI,” very few organizations are actually building the internal oversight to make it happen. Without a formal group in place, ethics always take a backseat to shipping the product and hitting revenue targets. There’s no one in the room with the authority to ask about biased training data, how a tool could be misused, or what the long-term impact on society might be before a product goes live. Those decisions are being made without any real ethical check.

This lack of oversight is why we keep seeing those awful headlines about biased AI, facial recognition that fails on non-white faces or hiring tools that filter out women. These problems usually come from skewed training data or just bad design choices that were never put through a proper ethical review. A real ethics committee, one with a mix of engineers, lawyers, ethicists, and social scientists, could spot these disasters before they happen, early in the design phase. Just telling your teams to “be ethical” is useless. You have to actually build the processes that force ethical principles into the work itself.

Transparency Black Box: Proprietary Models Dominate Commercial AI

Even as everyone screams for more transparency, the simple fact is that the vast majority of AI models in commercial use are proprietary black boxes. We can’t see how they work. This is a massive roadblock for governance. You can’t hold a system accountable if you have no idea how it reached its conclusions, and this opacity isn’t just about the code. It’s the training data, the testing methods, everything. Imagine a black-box AI that denies someone a loan. Without transparency, there’s no way to know if that decision was discriminatory and no real way to appeal it. The problem gets even worse with the big deep learning models, where sometimes (and this is the scary part) even the engineers who built them can’t fully trace why the model made a specific choice.

Companies always fall back on the same argument: intellectual property. They say opening up their models would give away trade secrets. Protecting IP is a valid concern, but it can’t be an excuse to put public trust and safety at risk. There are workable solutions here, like requiring third-party audits, creating standard formats for explaining model behavior, or putting the code for critical systems in escrow with a neutral third party. Without more transparency, any talk of oversight is just talk. It’s not real.

The Conventional Wisdom Misses the Multidisciplinary Mark

Most people think AI governance is just a job for lawyers or engineers. We hear it all the time, pass more laws, or build better bias-detection tools. Those things are part of the solution, but they’re not the whole picture. AI governance is a multidisciplinary problem, and it requires legal, ethical, technical, and economic thinking all at once. We spend too much time talking about what AI can do in a lab and not enough time on how to make it work safely in the real world. Passing a law is pointless if we don’t have the tech to audit for compliance, and it’s doubly pointless if all the market incentives still reward unethical behavior.

A technical fix for bias, like re-weighting a dataset, completely ignores the social reasons the data was biased to begin with. On the other hand, a law that just says “be fair” is unenforceable without a technical standard for what “fair” means and a way to measure it. To get this right, you need policymakers who get the tech’s limits, ethicists who can shape the design, economists who understand the market forces, and lawyers who can write rules that actually stick. It’s all connected. Any attempt to solve this from just one angle is doomed. We need these different experts working together from day one, not just having the technologists brief the lawyers after the fact.

Getting AI governance right is going to be hard, but we have to move fast and work together. We can’t let the tech get so far ahead of us that we lose control. The stakes for everyone are just too high.