AI Lab Workers Split Over Warnings of Mass Harm
Workers from OpenAI, Meta and DeepMind told the BBC they doubt that unchecked AI development will lead to tools able to kill people en masse.
The accounts came from text exchanges and conversations gathered by the BBC. All spoke on condition of anonymity because they were not permitted to speak to the press. Their identities are known to the BBC. Some answered recent high-profile warnings about AI risk with dismissal, including "Lol", "Haaaaaa" and "Bringing the luls".
The warnings include claims made last week by Jacob Coxon, a former Anthropic employee, that went viral and were echoed by others urging a slowdown in AI development. Coxon has said a group of AI agents — software that can take steps toward a goal without constant human input — based on models that do not currently exist could decide to create and then aim a biological weapon. That scenario centers on future agent systems, not on models now deployed.
Pushback has also been public. Colin Fraser, a data scientist at Meta, wrote on social media last week that there was no real evidence AI models would inevitably pursue a goal leading to human death. Rishub Jain, who founded the AI safety research firm Sampura Research this summer after seven years at DeepMind, was among those identified in the reporting around the debate.
The dispute follows public warnings about extinction risk. A public letter released in July was signed by 1,386 AI workers, according to The Spinoff, which on Sept. 15 published an article titled "Should we really be worried about AI ending humankind?" On Sept. 7, U.N. rights chief Volker Turk warned that artificial intelligence could pose an existential threat to humanity, according to Reuters. On Sept. 16, Reuters published a video titled "AI extinction risk: should we hit pause?" In it, technology reporter Aditya Soni covered the rise of "recursive self-improvement" or RSI, where a system improves its own design to get more capable. Industry leaders had earlier warned that AI poses a "Risk of Extinction," as reported by The New York Times.
The broader context here is a governance problem. International institutions are being asked to judge a risk claim when the technical community is divided. Turk's warning places AI inside the human rights architecture. The July letter and the viral Coxon claims press labs and governments to slow down. The anonymous rebuttals show that pressure is not shared evenly inside the labs.
Looking at what this means for diplomacy and regulation, internal dissent matters. Policymakers rely heavily on lab employees for early warning. If some see extinction scenarios as speculative and others see them as urgent, regulators face competing insider testimony. The question is less whether RSI or autonomous agent groups are possible in principle than what level of proof should trigger restraint. That line is now contested in public, in anonymous briefings, and through institutions from newsrooms to the United Nations.
