“Morally binding” is what you call a promise when you won’t print the terms

France 24 reports that President Trump met tech leaders at the White House and announced they’d signed a “morally binding” commitment to self‑regulate AI and “implement safeguards.”

That phrase is doing all the work — and none of it is auditable.

“Signed” wants to sound like an instrument. “Binding” wants to sound like enforcement. “Commitment” wants to sound like terms. But as reported here, there are no terms on the page.

A commitment you can test has basic handles:

• Who signed? Which companies, which officers.
• What did they sign? A named document, or at least a quoted clause.
• What safeguards? Not “safeguards” as a category — the actual practices.
• By when? A timeline is part of the promise.
• Verified how, by whom? Self‑policing is not verification.
• What happens if they don’t? A binding commitment without a consequence is just a vibe with stationery.

The same outlet can write in a keyed register when it wants to. In today’s report on pro‑Russian influence operations targeting French media, France 24 attributes the claim to a specific issuer (NewsGuard), supplies a quantified spike (“850% surge”), and describes the mechanism (fake stories impersonating real outlets).

On AI self‑regulation, the reporting stays at the level of posture: moral language substituting for operational language. That isn’t a complaint about “self‑regulation” as an idea. It’s a complaint about the way “binding” is being used to buy credibility without supplying the minimum that would let a reader check what was promised, by whom, and under what penalty.

If you can’t print the terms, don’t sell the bindingness.