The author runs a mid-sized AI infrastructure company.

I want regulation. The companies that operate in the AI infrastructure space need regulatory frameworks that establish clear expectations and that hold operators accountable for the consequences of what they build. The regulation we are getting, however, is calibrated for problems that are largely different from the problems the industry actually creates.

What the current regulatory conversation focuses on

The current regulatory conversation focuses on existential risks, on the long-arc questions of advanced systems' alignment with human interests, and on the broader frame of frontier-model behaviour. These are real concerns; they are not the concerns that produce the bulk of the day-to-day harms that AI deployment is creating.

What the actual harms look like

The actual harms that the day-to-day deployment is producing are more mundane and more measurable. Hiring tools that produce systematic bias against specific applicant categories. Customer-service tools that fail in ways that visibly harm specific customers. Content-moderation tools that exhibit the precision-recall trade-offs that make principled enforcement impossible at scale.

What useful regulation would address

Useful regulation would address the operational quality of these systems through requirements on testing, on transparency about failure modes, and on accountability for the harms the systems produce. None of this requires the kind of sweeping framework the broader conversation is converging on.

What the sweeping framework produces

The sweeping framework produces compliance burden that affects smaller operators disproportionately, that creates competitive advantages for the largest operators who can absorb the burden, and that distracts attention from the operational improvements the underlying systems actually need.

What I am asking for

What I am asking for is regulation that is calibrated to the actual harms, that produces testable requirements, and that accepts that the most consequential AI consequences are not the existential ones but the everyday ones. Building this regulation requires more granular technical engagement than the broader policy conversation has been willing to commit to.

The honest read

The honest read is that the policy conversation will, on past patterns, continue to focus on the more dramatic framings until something concrete forces it to engage with the operational realities. The practical consequence will be a regulatory framework that fails to address what most needs addressing while creating burdens that limit the broader ecosystem's healthier development.