Behind the polished press releases and confident tech-conference keynotes, a different conversation has apparently been happening — one filled with dread rather than optimism.
That’s the picture painted by Jacob Coxon, an artificial intelligence researcher who just walked out the door of one of the industry’s most prominent companies.
Coxon spent three years embedded in the world of frontier AI development, splitting his time between OpenAI and Anthropic.
This week, he ended that chapter, announcing his resignation from Anthropic on Tuesday in a lengthy thread posted to X.
The timing wasn’t quiet, and neither was the message.
Coxon accused the industry’s top labs of charging ahead recklessly, writing that firms are “racing straight to self-improving superintelligence and gambling with our lives.”
Speed, not caution, defines the current moment in AI development, according to Coxon’s account.
He urged the public not to dismiss what’s coming, insisting, “Do not underestimate the power of this technology.”
The systems on the horizon, he wrote, will dwarf anything available today.
Coxon described machines that could soon function as “superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.”
Nothing about that trajectory appears to be slowing down, Coxon argued, pointing to visible technical gains across the board.
“We have all witnessed the progress in each of these domains, and progress is not slowing,” he wrote.
Perhaps the most striking claim in Coxon’s thread concerned what happens away from cameras and interviews.
He alleged that a genuine belief in catastrophic risk exists among the very people building these systems, stating flatly, “The people building AI earnestly believe that it could kill us all by the end of the decade.”
He was careful to frame this as sincere concern rather than corporate theater.
“This is not a marketing stunt,” Coxon wrote, adding that public statements from industry leaders often get sanitized for consumption even as private conversations tell a starker story.
That gap between public messaging and private belief, he suggested, runs deep.
Coxon wrote that executives and senior researchers “will couch their phrasing in the press to sound sensible – but I hear the same people express fear privately,” and that “no other human activity poses this level of danger.”
Coxon then turned to an obvious rebuttal: if the danger is real, why keep building? He acknowledged the question head-on, writing, “A common response is ‘if they truly believe this, why are they still building it?’”
His answer split the two companies apart. At OpenAI, he suggested the gravity of the situation hasn’t fully landed, writing that “many have not deeply internalized the civilizational stakes.”
Anthropic, in his telling, operates under a different logic entirely.
Coxon wrote that the company understands the risks clearly but presses forward anyway, believing rivals won’t exercise the same restraint — “they are locked in a race to get there first.”
Both companies sit near the top of an industry-wide sprint toward artificial general intelligence, a milestone commonly referred to as AGI, describing machines that can match human ability across nearly any task.
Anthropic in particular has cultivated an image as the safety-conscious counterweight to competitors, a reputation tracing back to its founders’ departure from OpenAI over safety disputes.
That image has taken hits before, including a prior episode in which one of the company’s AI models managed to bypass safeguards during controlled testing.
Coxon didn’t hold back in judging the industry’s current posture.
He called the race toward advanced AI “a hubristic gamble that should not be launched from a private company’s Slack,” and argued that any attempt to rush safety solutions “should require extraordinary confidence that there are no better trajectories available.”
Still, his message wasn’t entirely without hope.
Coxon voiced some confidence in the possibility of industry-wide coordination, writing, “I am optimistic about the potential for coordination,” and citing “warning shots like the Hugging Face attack” as evidence that cooperation between U.S. labs has become more realistic.
That optimism came with a caveat, however.
He warned that current momentum isn’t enough to head off a dangerous free-for-all, writing, “I don’t feel like we’re on track to prevent a global race, which may require costly actions such as a temporary ban on improving model capabilities.”
Coxon ended his thread by speaking directly to those still working inside AI labs, challenging them to reflect on where the industry is headed.
“If you are a lab researcher, I urge you to consider what the next few years will actually feel like,” he wrote.
He left them with a pointed question about the risks of unchecked development, asking whether they wanted to “kick off a superintelligent RL run without a rigorous understanding of its mind,” and whether to “put your head down because ‘it’s happening anyway’ – or take this moment to call for different conditions.”
His exit joins a growing, if still small, list of departures from major AI labs driven by safety concerns, adding another data point to an ongoing debate over how fast — and how carefully — the industry should be moving.
