Opinion: Government Clearance for AI Models Sets a Precedent We Are Accepting Too Quietly

Government clearance for AI models is now a fact of American life, and almost nobody blinked. OpenAI released its GPT-5.6 family publicly only after the US Commerce Department finished weeks of testing, CNBC reported. For two weeks before that, the models went only to a “small group of trusted partners” at Washington’s request. This is a column, and its argument is simple: we should be paying far more attention.

Start with what is reasonable. Frontier models carry real risks in biology and cybersecurity. Pre-release evaluation by a standards body is a defensible idea. Democracies inspect aircraft before they fly.

Why government clearance deserves scrutiny anyway

But notice what has actually been built. A June executive order asks developers to “voluntarily” hand cutting-edge models to the government before release. Voluntary, when your regulator asks, is a word doing heavy lifting. The same administration has floated taking a 5% stake in OpenAI. Reviewer, shareholder and customer are becoming the same entity.

Software is also speech, in law and in practice. A release gate on models is a release gate on the most powerful publishing tool ever built. Today the gate checks for bioweapon uplift. Tomorrow’s administration decides what else it checks for.

The test of a good precedent

Ask one question of any new power: would you trust your least favourite future president with it? If the answer is no, the power needs statutory limits, transparency and appeal rights. None of those currently exist here. There is no law, only leverage.

AI safety review may well be necessary. That is exactly why it must be legislated, bounded and public, not improvised through favours between one company and one government. Guardrails for the machines are welcome. Guardrails for the gatekeepers are overdue.

The international contrast sharpens the point. In Geneva, the UN spent the week debating AI rules in public, with every government’s position on the record. Washington’s model review happened in private, between one agency and one company. Public rules can be contested. Private leverage cannot.

Fact basis: CNBC (July 8, 2026), Nextgov/FCW, and The Next Web reporting on the GPT-5.6 clearance process and the June 2026 AI executive order. This is an opinion column; the judgments are the author’s.