The AI Safety Vacuum at the Top of Washington | Column

Opinion / Column. The person running America’s AI safety agency has resigned after just three months. The departure lands as models grow faster and cheaper every week. It leaves a gap at exactly the wrong moment. Someone should be minding this risk.

Three months is not a tenure. It is barely an induction. Whatever the personal reasons, the signal is unmistakable. The job looks unworkable from the inside.

Why the AI safety gap matters

AI safety work depends on continuity and standing. Regulators need time to build technical depth. They need authority to demand real answers from labs. Constant churn destroys both.

The timing is poor. Governments met this month at the Global Dialogue on AI Governance. They wrestled with rules for a technology outrunning them. Leadership vacancies weaken any seat at that table.

Industry will not pause while Washington reshuffles. Firms are shipping specialised models and racing on cost. Chip partnerships are redrawing the hardware map. The frontier moves whether or not officials keep up.

Some argue the agency was toothless anyway. That criticism has force. Voluntary commitments rarely bind anyone. A weak office losing its head changes little in practice.

Yet the alternative is worse. Without any public capacity to evaluate systems, the only assessments come from the companies selling them. Self-assessment is not oversight. It never has been.

Freethinkers should resist both extremes here. Panic serves nobody. Nor does blind faith in corporate goodwill. The sensible demand is boring and specific: fund the office, staff it, and give it real powers.

We have argued before that voluntary standards do too little. A vacant chair does even less.

Appoint someone. Then let them stay long enough to matter.

Fact basis: Reporting on the resignation via Artificial Intelligence News and UN News coverage of the Global Dialogue on AI Governance, July 2026. This is an opinion column.