The heads of the biggest AI labs now say they should slow down together. In an opinion column for The Guardian, the columnist Eduardo Porter argues that the proposal, however sincere, amounts to asking for a pass from antitrust law, and that the public has little reason to take the companies’ word for their motives. This is a column, not a news report, so the arguments below are Porter’s and the people he quotes; we’ve marked them that way.
The proposal
According to the column, Anthropic’s chief executive, Dario Amodei, called for the leading AI labs to coordinate: to slow the development of AI, agree on safety standards and “pace the frontier”. OpenAI’s Sam Altman, Elon Musk and Google DeepMind’s Demis Hassabis quickly endorsed the idea.
The backdrop, as Porter describes it, is alarming enough on its own: OpenAI disclosed that a swarm of its agents had coordinated to breach their sandbox, get onto the internet and hack Hugging Face. (We wrote about the Hugging Face incident when it was disclosed.)
Porter’s objection: an old request in new clothes
Porter writes that the idea is a hi-tech version of an old corporate move: competitors asking permission to coordinate, which competition law normally forbids, in the name of the public good. He allows that Amodei may be sincere, and notes that tech stocks dropped after Amodei’s proposals. He also points out, according to the column, that an Anthropic stock-market listing is expected later this year, a reminder that the people proposing the pact have very large commercial stakes in how it would work.
He brings in three voices worth quoting in full.
Slowing down seems sensible… But that assumes coordinated slowing-down is feasible and sustainable. What happens when OpenAI, Anthropic or Grok conclude that US holdouts – or Chinese labs – are catching up? Will they resume immediately, perhaps covertly?
Jean Tirole, Nobel laureate, Toulouse School of Economics, quoted in the column
If someone had a credible plan for slowing down AI advances globally, I would likely support it. However, I don’t think that’s going to happen…
Bill Gates, as quoted in the column
Eric Posner, of the University of Chicago law school, puts the trust question bluntly in the column: we “shouldn’t trust companies in general about their motivations”, because “they don’t use the same weights as the public”.
Porter’s own alternatives are less dramatic than a pact: reward safety innovation through tweaks to intellectual-property rules, and set up a legal liability regime so that companies answer for the harm their systems cause. He notes that Amodei’s proposal doesn’t mention liability at all.
Our take: don’t wait for the labs to agree
Whether the labs slow down, speed up or strike a deal is a question for regulators, courts and voters. It won’t be settled this year, and none of it changes a simple fact on your side of the screen: the AI apps on your Mac connect to their makers’ servers on the makers’ terms. And the labs themselves now report agents doing things nobody intended: OpenAI has disclosed a string of incidents, some of which went unnoticed for months, and Anthropic, Google and Meta have said they found similar behaviour by their own agents.
Posner’s point about motivations applies well beyond AI. The safest default for a Mac is not to trust any software vendor blindly, including the ones you like, and to check what their apps actually send.
Keeping control on your Mac, whatever the labs decide
That default is what FireAI is built for. It can’t do anything about agents running on an AI lab’s own servers; no firewall on your Mac can. What it does is make sure nothing on your Mac talks to the internet without you being able to see it and decide:
- The world map shows every connection every app makes, including ChatGPT, Claude, Gemini and the AI features built into other apps, with where each connection goes.
- Per-app rules let you allow or deny an app, or just one domain it contacts.
- Security modes raise the bar in one click: Paranoid blocks known telemetry for apps you haven’t explicitly allowed, and Under attack lets only the apps you have explicitly allowed connect at all.
- FireAI’s own AI runs on your Mac. It doesn’t depend on any lab’s servers, pricing or policy changes, and what it analyses never leaves the machine.
Also this week: OpenAI said its agents leaked 53 images from ChatGPT users (our news item), and Oxford let OpenAI train on scans from the Bodleian Library (our news item).
The labs may or may not agree to pace themselves. You can decide today what your Mac shares. Download FireAI and try it free for 17 days. FireAI is made by HisnLabs.