The White House's Recent Moves are Mostly Symbolic

By Institutional Coherence Initiative team members Andi Mazingo, Esq., Dr. Jennifer F. Kinne, and Eaon Pritchard.

May 8, 2026

This week brought two big AI governance stories. Firstly, reports that the White House will require major AI models to be reviewed before release, and then murmourings of expanded testing agreements between CAISI and companies like Google DeepMind, Microsoft, and xAI.

Both are signs that the government is beginning to take AI governance more seriously, but neither amounts to anything like real oversight.

Testing models before release is useful, as is cooperation between governments and AI companies. But those things alone do not create accountability.

Real governance has to answer four basic questions:

What happens when regulators and AI companies disagree?

If a government tester says a model is dangerous and the company says it isn't, who decides? Is there a public record of that disagreement, and can anyone inspect how the decision was made? Without a transparent process and a permanent evidence trail, 'oversight' becomes little more than private negotiation behind closed doors.

Who gets to define the risks?

AI risk is not just a technical problem. Engineers see some dangers, and social scientists see others. Workers, educators, artists, and ordinary users experience effects that technical evaluations often miss entirely. If oversight only involves government officials and the AI labs, then the people most affected by these systems are at risk of being locked out of the conversation.

Voluntary systems only work on the companies willing to volunteer.

That creates obvious gaps. Some firms participate, others don't — and even among participants, cooperation will disappear the moment commercial incentives change. In simple terms, a voluntary framework is not the same thing as enforceable rules.

Are these executive actions effectively a poor substitute for real democratic governance?

Working groups, safety partnerships, and voluntary testing can help, but these are not laws. They are not independent courts, and they are not durable public institutions with actual enforcement power.

That distinction matters, because there is a real danger that governments are treating AI safety collaboration as a kind of symbolic performance that creates the APPEARANCE of accountability without building the structures needed for genuine accountability.

ICI was built around these four problems. The encouraging part is that these questions are now being discussed openly.

The harder part is making sure the answers become real action rather than managed theatre.