Nvidia CEO Jensen Huang has weighed in directly on Jacob Coxon, the pretraining researcher whose resignation from Anthropic and viral warning about existential AI risk have dominated AI conversation this week.
According to a post from venture investor Brad Gerstner, Huang described Coxon’s comments as outlandish and deeply untrue. Huang reportedly said he has respect for the work being done at the frontier labs, but that Coxon’s remarks were wrong, arrogant, and dismissive of the amount of safety work already happening across the industry.
The pushback lands just as Coxon’s resignation thread has turned into one of the most talked-about moments in AI this year. Coxon, who said he spent three years on pretraining research across both OpenAI and Anthropic, argued that people actually building frontier systems privately believe the technology could pose an existential threat before the end of the decade, even as they soften that message in public. He was especially pointed about Anthropic, saying the company understands the risks internally but has convinced itself it has no choice but to keep racing, reasoning that a less careful competitor will get there first if it doesn’t.
Huang’s response fits a pattern he has leaned into for months. Speaking at a Goldman Sachs conference, he argued that much of the current alarm around AI is being manufactured by an industry with something to sell, remarking that creating a problem is one of the best ways to create demand for a product. He made similar points in a July interview, dismissing claims that AI could end humanity or wipe out huge swaths of the American job market as baseless. ““The fact that this is going to be the end of humanity — it’s complete nonsense. The fact that this is going to destroy half of the American jobs is complete nonsense,” he said.
Coxon’s post has also drawn scrutiny of a different kind. A separate wave of posts has raised questions about how organically the thread spread, pointing to its speed, the timing of a Wall Street Journal piece that reportedly went up minutes earlier, and Coxon’s past ties to AI-safety-adjacent funders and fellowships. None of that has been confirmed as coordinated activity, and it hasn’t stopped others inside the industry from backing Coxon’s underlying point — Anthropic’s own alignment science lead, Evan Hubinger, has said he sees more than a 10 percent chance that AI could kill everyone within the next decade, a comment that complicates any attempt to wave the episode away entirely.
Huang, for his part, has been consistent in arguing that the loudest doom narratives serve the commercial interests of the people telling them, whether that’s a cybersecurity industry gearing up new products or AI labs positioning themselves as the only ones responsible enough to keep building. Coxon has not publicly responded to Huang’s comments, and Anthropic has not issued a statement.