NEW YORK, September 10 — The AI industry knows what it may be building, and it is building it anyway. That is the warning at the center of a PBS NewsHour segment that aired Wednesday, September 9, titled “AI researcher warns companies are ignoring catastrophic risks,” reported by Geoff Bennett with Azhar Merchant. The researcher is Jacob Coxon, who has worked at both Anthropic and OpenAI.
In a written statement quoted in the segment, Coxon did not soften his language: “The people building A.I. earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.
If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible. But I hear the same people express fear privately. No other human activity poses this level of danger.” What made the warning land, according to The Wall Street Journal’s Amrith Ramkumar, who was interviewed on the broadcast, is partly who delivered it.
Coxon is quitting the industry, Ramkumar said, and therefore “no longer has skin in the game.” He also carried the credibility of having worked at two of the top AI companies.
A Timeline That Is Not Distant
Ramkumar told PBS that Coxon discussed major risks arising “late next year or the year after,” with the potential extinction of humans following after that depending on how events unfold. The severity of the claims, Coxon’s departure, and several incidents of AI systems hacking other companies combined, in Ramkumar’s words, to create “a pretty volatile moment.” Asked to unpack the roughly 10 percent extinction estimate that has circulated around these warnings, Ramkumar said it does not refer to a chatbot out to kill everyone.
It refers to a theoretical chain of events involving a much more capable future AI system. The key concern, he said, is “recursive self-improvement” — when a model can improve and train itself, and other AI systems then become much more powerful.
People say that could be hit within a couple of years. After that, Ramkumar said, it is very hard to predict what could happen. He described scenarios in which models train themselves and learn to reject human commands.
Once able to do so, they could command drone swarms or robots and might decide humans are not necessary to fulfill their objectives. Ramkumar stressed that the researchers assign probabilities to these outcomes and are not necessarily saying the probability is high — but that it is on the table and possible.
Skepticism Has Followed These Alarms Before
The segment also noted that there has long been skepticism about the motives of CEOs and researchers who raise alarms about the potential future of AI. That tension is part of why the current moment stands out. The broadcast carried a date of September 9, 2026, at 6:55 PM EDT.
What to watch next is whether the pressure Coxon and others are placing on government to get involved translates into anything concrete — or whether, as he warned, the race between companies and with China keeps the private sector moving ahead regardless.
