Global Edition
The Doom Ledger
Est. 2026
AI is not a new church, and people don’t need a new pope.

Open Letters and Pledges: A Year of Safety Theatre

Statements of concern are cheap to sign and easy to ignore. Measuring whether any of them changed a release decision is the part almost nobody does.

Microphones at a press conference — photo by Matti Blume, licensed under CC BY-SA 4.0 via Wikimedia Commons.

The last few years produced a remarkable volume of collective statements about artificial intelligence risk. Researchers signed letters. Chief executives published principles. Companies adopted charters. Several governments issued declarations.

The question that follows every one of them is uncomfortable: did anything change?

Vocabulary, attribution, and what counts as respectable

It would be a mistake to conclude that these documents do nothing. They establish vocabulary. They make a position publicly attributable, so that a later reversal is visible. They give regulators something to reference, and they shift what counts as a respectable position inside companies.

The mistake is to treat them as commitments with teeth. A pledge to pause development of something nobody is close to building costs nothing. A principle that no business decision has ever been weighed against is decoration.

Advertisementin-article · responsiveAfter the opening section of a long article. Never between a heading and its own body.

Four questions to ask of any pledge

  • Does the commitment specify a trigger, or only a sentiment?
  • Is there a named person accountable for it?
  • Would violating it be visible from outside the company?
  • Is there any cost to non-compliance?

Nobody can afford to move first on restraint

The theatre is not primarily a product of insincerity. It is a product of a competitive environment in which no participant can afford to move first on restraint, and in which public concern is real enough that saying nothing is also costly. Declarations are what an industry produces when the incentives point away from action but the audience expects a response.

The remedy is not fewer declarations. It is declarations with triggers, auditors and consequences — which is to say, regulation.

Thresholds that staff cited internally

Not every declaration was empty. Some companies published specific thresholds and pre-commitments that later constrained their own release decisions, and were cited by staff internally when a product team wanted to move faster.

Advertisementin-article-2 · responsiveRoughly two thirds down a long article.

The pattern that separated the effective from the decorative was specificity. A commitment naming a capability, a safeguard and a decision-maker created something an organisation could be held to. A commitment expressing concern created nothing.

Written for four audiences at once

Declarations are often written for several audiences at once: employees, investors, regulators and the public. The resulting language is broad enough that each can read what it wants, which is precisely what makes it unenforceable.

The remedy is to write for one audience at a time. A document aimed at a regulator should be specific and checkable even at the cost of being unflattering, and most industry documents are not written that way because the same text has to work in a press release.

Image credit and licence details for every photograph on this site are listed on the credits page. This article is editorial content; it carries no sponsored material.

Related

Advertisementfooter-banner · 970x90End of page, above the site footer. Never inside the footer itself.