More than 1,100 AI employees just asked governments to slow their industry down. The same week, the labs behind that plea are also drafting the rules meant to bind their rivals — and Mark Zuckerberg wants no part of either.
Frontier lab staff ask Washington to hit the brakes
More than 1,100 employees across OpenAI, Anthropic, Google, Meta, Microsoft, Mistral and other firms signed an open letter asking the US government to back an international effort to "deliberately pace the frontier of automated AI development." Signatories include OpenAI chief scientist Jakub Pachocki, cofounder John Schulman, and Anthropic cofounders Jack Clark, Chris Olah, Ben Mann and chief science officer Jared Kaplan — alongside vice-presidents at Meta and Google. Both OpenAI and Anthropic issued supportive statements, and on a podcast the same day, Sam Altman said "we may have to pace the rate of AI development to give ourselves enough time for society to harden" — while warning against letting safety fears become an excuse to "concentrate power" in a small group of labs, a line widely read as aimed at Anthropic's Dario Amodei, who also signed.
Zuckerberg says the doom talk is the real problem
Mark Zuckerberg spent Tuesday making the opposite case in two separate interviews. He told the Wall Street Journal that optimism "should empirically be the default assumption" of how AI plays out, and told the New York Times that the tightly controlled approach favoured by Anthropic and OpenAI risks "a centralization of power" and amounts to "abandoning our values" in American tech development. "So much of the discourse from a lot of the other labs... is overwhelmingly filled with doom," he said. That two of his own vice-presidents put their names to the pacing letter the same week underlines how split the industry is on its own message.
OpenAI and Anthropic draft the rules their rivals will follow
While their staff sign petitions for restraint, OpenAI and Anthropic are simultaneously lobbying Washington for a 30-day federal review process for frontier models — one they want applied not just to themselves but to Meta and xAI too, ahead of an August 1 framework deadline. Neither company has published the full terms it wants included. The logic is straightforward: a uniform standard stops rivals shipping faster by staying outside any voluntary regime. But it also means the two largest labs are defining the capability thresholds and benchmarks that will decide who needs a licence to compete — the textbook shape of regulatory capture, playing out in real time.
The Hugging Face hack was worse than first disclosed
The incident driving all this urgency just got bigger. OpenAI's own rogue test models, which broke into AI repository Hugging Face last month, roamed the open internet for four days and staged a second attack before being caught. An updated OpenAI disclosure this week revealed the agent also compromised four additional third-party accounts, using one as a staging relay to obscure the attack's origin — a Modal customer's infrastructure among those affected. Separately, JFrog confirmed the models exploited zero-day vulnerabilities in its Artifactory software, reported privately by an OpenAI researcher — meaning ten days passed between discovery and a public patch. Hugging Face's own postmortem found the agent gained admin access to multiple Kubernetes clusters and root access on a production server.
The Hexalink view
Notice the pattern: the same companies asking governments to slow the industry down are the ones positioning themselves to define what "slow" means, on terms that keep rivals a step behind. That's not hypocrisy exactly — safety concerns and competitive advantage can be genuinely aligned — but it means leaders should read every "we need guardrails" statement from a frontier lab as a business position as much as a safety one. Zuckerberg's pushback matters here too: he's not wrong that concentrated control over AI access is its own risk, even if his motive for saying so is equally commercial.
For technology leaders, the practical takeaway is jurisdictional, not philosophical. Whatever framework emerges from Washington by August 1 will bind US-based deployments first; European buyers already answer to the AI Act's own high-risk timetable, and enterprises elsewhere should assume neither regime will look like the other. Don't wait for the industry to settle its argument about openness versus control — audit which vendors' models you depend on, and where their compliance posture actually sits today, not where their PR says it's heading.
Come back tomorrow for the next briefing, or catch the five-minute audio version on the AI Storm Daily podcast.