Artificial intelligence turned up at the United Nations Security Council. Sam Altman of OpenAI, Dario Amodei of Anthropic and Clément Delangue of Hugging Face addressed the fifteen members on the development and the risks of the technology.
Pause on the venue before the quotations. The Security Council is the body that deals with armed conflict, sanctions and non proliferation. It has the power to impose binding obligations on states. Three private companies were there to describe the dangers of a technology they build, sell and set the release schedule for. That is not a scandal. It is simply worth saying out loud, because the quoted remarks read differently once you have.
The promise with no unit of measurement
Amodei, speaking by video link, named two risks: misuse for the creation of biological weapons by terrorists, and loss of control. He called for collaboration between governments around the world and listed ways to contain the danger, including universal standards and global safety systems. No leader, no company and no country, he said, can face this challenge alone. And he added that his company would slow down as much as necessary to guarantee that AI is safe.
We will slow down as much as necessary.
First person plural. Future tense. The subject of the verb is the company, and the sentence names nobody else. No stated threshold, no unit in which the slowing would be measured, no published trigger, and in those words no verifier other than the firm making the judgement.
This is not an accusation of bad faith. Take the sentence as entirely sincere and it still does no work, because a commitment whose only guarantor is the party making it is not a commitment. It is an intention. Intentions survive exactly as long as the competitive conditions that make them affordable.
And it was said to a body that can impose binding obligations on states, in the form of a voluntary undertaking. States and national authorities can impose limits too, of course. The undertaking quoted here did not come from any of them.
Weather reports about decisions
Altman put it in the register that has become standard for this conversation. We have a choice in front of us, he said: artificial intelligence can be a new Renaissance and can expand knowledge, or it can move faster than our institutions, and attention is required.
It is a graceful formulation and it is slightly meteorological. AI does not move. It is moved. The pace at which capability is released is an industrial decision with a name, a board and a quarterly logic behind it, not a weather front arriving off the Atlantic.
Once the speed is described as a property of the technology rather than a choice by its producers, the only remaining question is how institutions should adapt to it. Which is a considerably more comfortable question than asking who set it.
He also said that AI must be built for people and must put people at the centre, and that individuals need to be empowered at the individual level.
Individual responsibility is the one commodity this industry distributes generously, always downwards. The user who approves the wrong permission. The customer who should read the notices. The citizen who should be AI literate. Meanwhile the decisions that determine what can be approved, what is disclosed and how fast the next system ships stay where they were.
What is actually being asked
Strip the language and the substance of the appearance is coherent: the companies want universal standards and international cooperation, and they say the challenge exceeds any single actor. They are right about that. The risks named are not fictional, and a firm publicly identifying loss of control as one of its two principal concerns is saying something few industries have ever said about their own product.
But there is a difference between asking for standards and accepting constraints. Standards, in the version these companies advocate, measure and describe. Constraints stop things. Universal standards, global safety systems, collaboration between governments: what was quoted from that session sits in the first category, and the Security Council is a body built for the second.
Three executives explained what their technology might do to the world. The obvious follow up, what happens if they judge wrongly and who is entitled to decide instead of them, is not answered in any of the remarks that were quoted.
We will slow down as much as necessary. Necessary according to whom, verified by whom, and enforced by whom if the answer is inconvenient.
Raffaele Di Marzio
All my “insane” books on cybersecurity and governance are here 👇 https://cyberium.limited/bookshelf.html