Political Tech
When the most powerful AI company in the world gets to define what counts as dangerous, the rest of us are working from its rulebook.
NewsOnScale Staff
September 1, 2026
There is a particular kind of power that doesn't announce itself. It doesn't come with a press release that says 'we are now setting the terms of acceptable reality.' It comes dressed as a policy document, a governance framework, a responsible AI plan. And this week, Google published one.
The company's AI governance framework, drawing attention from Tech Policy Press and policy watchers across the spectrum, lays out how Google intends to define, identify, and respond to harm caused by its AI systems. On the surface, this looks like corporate accountability in action. A major player acknowledging risk, putting guardrails on paper, signaling seriousness to regulators and the public alike. That framing is not entirely wrong.
But it is dangerously incomplete.
## The Definition Problem
The most consequential part of any harm framework isn't the enforcement mechanisms — it's the definitions. What counts as harm? Who is considered a potential victim? Which harms are treated as acceptable tradeoffs against utility or profit, and which cross a bright line?
These are not technical questions. They are political and ethical ones, and they have always been contested in democratic societies through legislation, litigation, and public deliberation. When Google answers them internally — even thoughtfully, even with genuine intent — it is making governance decisions that would normally require some form of public accountability.
The Brookings Institution made a related point this week, publishing a direct call for Congress to pass federal AI legislation rather than allow this vacuum to persist. Their argument is not new, but the urgency behind it is sharpening: every month that federal standards remain absent is another month in which private companies fill the gap with their own.
## Preemption in Reverse
The Regulatory Review flagged a parallel concern this week around regulatory preemption — traditionally, the question of whether federal rules should override state ones. But there is an underexamined version of preemption happening right now in the opposite direction: industry self-regulation is quietly preempting government regulation before government regulation even exists.
When a company the size of Google publishes a detailed governance architecture, it does not just describe how that company will behave. It shapes what policymakers and the press treat as the reference standard. Congressional staffers read these documents. Regulators cite them. Journalists use them to evaluate other companies' practices. The Overton window for what 'responsible AI' looks like shifts accordingly.
This is not a conspiracy. It is how institutional power operates in information-rich environments. And it puts the burden of scrutiny squarely on those of us covering this space to ask harder questions than the framework itself invites.
## What the Framework Doesn't Tell Us
Google's governance plan, as described in coverage so far, draws boundaries around harm categories — but published corporate frameworks almost never answer the questions that matter most. How are edge cases adjudicated, and by whom? What recourse exists for users or communities who believe they've been harmed in ways the framework doesn't recognize? Are there external auditors with genuine independence, or does verification remain internal?
These gaps aren't necessarily signs of bad faith. They may simply reflect the limits of what any single organization can specify in a public document. But they are precisely why self-governance cannot be a substitute for public law.
## The Stakes Are Structural
For readers of this publication, the concern isn't limited to AI safety in the abstract. It extends to the agent economy specifically — to systems that take actions on behalf of users, intermediaries that shape what information reaches whom, and platforms that may suppress or amplify civic information in ways that are governed, if at all, by frameworks exactly like this one.
When the rules are written by the same entities that profit from the systems those rules govern, accountability becomes a performance rather than a mechanism. Google's framework may be well-constructed. It may even be well-intentioned. But good intentions are not a governance structure.
Congress has the authority to change that. So far, it has chosen not to use it.