Invest1 publisher3 min readPublished
Altman says OpenAI's investors were warned safety might cost them money
Altman told Fortune that OpenAI will not go public this year and that its backers were warned safety work could cost them. The coordinated slowdown he and Amodei want still needs an antitrust exemption to work.
The Investor · Invest desk
What happened
- Warnings from staff resigning from, and still working at, Anthropic, OpenAI and Google DeepMind that the leading labs are risking human extinction ran through the global news cycle for a week, Fortune said.
- Altman told Fortune's Alyson Shontell that OpenAI favours coordinating an industry-wide slowdown with Anthropic, SpaceX, Google DeepMind and Meta, and hinted that talks were already underway.
- Amodei's blog post called for a coordinated slowdown or pause among frontier labs in democratic countries and said some of that coordination would require an antitrust exemption from the government.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint The exemption clause hands a private agreement's fate to the government, so the labs cannot run the core of their own plan without the administration Fortune says resists a slowdown.
- cost OpenAI's existing investors were told going in that safety work might cost them money; buyers in any IPO would be pricing that commitment for the first time, without the same warning.
- exposure If half of companies are ignoring their own AI governance policies, those self-written documents become the readiest evidence against them in a dispute.
- precedent With a single state statute the only law in play, the next argument over frontier-lab conduct is likelier to be fought in California than in Congress.
A pause that requires an antitrust exemption is not a document five companies can sign among themselves. Amodei put the condition in writing: coordinating with other labs would in some cases need an exemption from the government [6]. Fortune's headline on the week says Trump resists calls for a slowdown [1]. The one part of the plan with legal force depends on the party least inclined to grant it.
Altman's version arrives with a cost attached, or rather with a cost his investors have already agreed to. He said he would have no problem telling investors that OpenAI had taken actions to prioritise safety that had cost them financially, and that OpenAI's investors were warned of that possibility going in [4]. He also said OpenAI would not go public this year, partly because of current concerns about the safety of the latest models and partly, he hinted, because the business isn't yet in the right place [5]. A private company can absorb a safety-driven revenue hit in a letter to holders who were told to expect it.
Anthropic is nearer a listing, by Fortune's account, with OpenAI edging closer behind it [11]. That changes who pays. A commitment to slow down, made before an IPO, has to be described in a prospectus and priced by buyers doing it for the first time. Amodei's other pledges are cheaper to keep: independent evaluators permanently on site, with the nonprofit METR named as his preferred partner [7], and an international governance agreement with China and other authoritarian states, if possible [8]. The second is a request to governments.
Before this week the political argument about AI ran on data centre construction and utility bills, jobs, education, mental health, algorithmic discrimination and civil liberties [16]. Extinction risk is now inside that argument, carried by resigning and serving staff at Anthropic, OpenAI and Google DeepMind [2]. It has produced, so far, one item with legal force. Fortune's edition lists six things [14]; exactly one names a law already on the books, California's AI safety statute, which a watchdog group says OpenAI is violating [12][15]. The other enforceable-looking thing is self-written: E&Y found half of companies aren't following their own AI governance policies [13].
I would expect a plaintiff or a state regulator to quote a company's own governance policy before quoting any federal standard, since the list does not contain one [15]. Two developments would change that. An antitrust exemption granted to the labs would make the slowdown a matter of federal permission [6], and a signed agreement with China would move it onto treaty ground [8]. Without either, next year's compliance pressure comes from California and from documents the companies wrote themselves.
What to watch
- Whether the administration grants the antitrust exemption Amodei said lab coordination would in some cases require, and on what conditions.
- Whether any Anthropic listing document describes a slowdown or pause commitment as a risk factor for new buyers.
- Whether any enforcement follows the watchdog group's claim that OpenAI is violating California's new AI safety law.