The people writing AI laws don't understand AI, and the people deploying AI in boardrooms don't either — so who's actually steering this thing?
The Summary
- Democratic Rep. Don Beyer, co-chair of the Congressional AI Caucus, says recent AI incidents demand federal safeguards including mandatory reporting for advanced model failures
- Nearly three-quarters of corporate directors admit they lack the skills to understand AI issues, even as they greenlight AI deployments
- Congress wants a hybrid model: federal oversight combined with technical expertise from AI companies themselves
- The expertise gap exists at both ends — lawmakers crafting rules and executives signing checks both flying blind
The Signal
We're watching a regulatory scramble in real time. Rep. Beyer's call for safety testing requirements and mandatory incident reporting follows what he vaguely terms "recent AI incidents" — no specifics given, which is either careful messaging or a tell that even the AI Caucus co-chair doesn't have full visibility into what's breaking. His proposed framework pairs federal regulators with technical staff from AI companies, a structure that sounds reasonable until you ask: which companies, chosen how, accountable to whom?
The timing matters. Beyer's pushing this just as boardroom data reveals 73% of directors feel unprepared to evaluate AI risks. These aren't junior managers. These are the fiduciaries approving AI budgets, signing off on agent deployments, and betting company futures on models they openly admit they don't understand. If the people with legal responsibility for corporate AI strategy are this far behind, what does that say about the actual deployment decisions happening three levels down?
"Congress should create a regulatory structure that combines federal oversight with technical expertise from AI companies."
This hybrid model has a built-in problem: the technical experts work for the companies being regulated. It's not a new tension, but it's sharper with AI because the knowledge asymmetry is wider. You can't borrow a nuclear engineer from Westinghouse to help write reactor safety rules and pretend there's no conflict. Same principle applies when the only people who truly understand frontier model behavior are the people building and selling them.
Key structural tensions:
- Lawmakers lack technical depth to write effective rules
- Corporate boards lack AI literacy to evaluate risk
- AI companies have expertise but obvious incentive conflicts
- "Recent incidents" remain unspecified but serious enough to move Congress
The Implication
If you're building with AI agents right now, assume the regulatory ground is about to shift. Mandatory incident reporting sounds bland until it's your production system that failed and you're deciding what counts as reportable. Start thinking like you'll need an audit trail. Document model behavior. Know what your agents are doing and why.
For everyone else: the expertise gap at the top means decisions about AI deployment are being made on vibes, not analysis. That creates opportunity for people who actually understand this stuff, but also means a lot of expensive mistakes are incoming. When three-quarters of directors are winging it, you get both regulatory overcorrection and preventable failures. Watch for both.