The UN General Assembly opened its 81st session with concerns about AI governance as a central theme in the speeches of heads of state and government, alongside war and climate change. Secretary-General António Guterres set the tone for the debate on the first day, warning: “Artificial intelligence must be shaped by humanity, with humanity, and for all of humanity.” He added that “decisions of life and death must never be left in the hands of machines” and that “killer robots must have no place in our future,” referring to autonomous weapons. He also noted that the developers of these systems themselves are now “raising the alarm” about a development that could be spiraling out of control, and proposed an international risk management framework that places greater responsibility on the powers with the most advanced capabilities, particularly the United States and China.
On September 9, Jacob Coxon, an AI researcher who had worked at both OpenAI and Anthropic, posted a thread on X that garnered over 115 million views in just a few days, essentially stating that “the people building AI seriously believe it could kill us all before the end of the decade.” He accused both labs of “playing with our lives” by racing to be the first to achieve superintelligence without adequate safety measures. He warned that self-improving systems could soon surpass human control, acquire resources autonomously, and become, in practice, uncontrollable.
The following day, the situation escalated when Evan Hubinger, head of Alignment Science at Anthropic, publicly backed him up: “Jacob has a point here: we genuinely, in all seriousness, believe that AI could kill all of humanity.” He even put it into concrete numbers: a greater than 10% probability of human extinction in the next decade. And he added something even more worrying: Anthropic “still doesn’t have a plan to resolve the superintelligence alignment.”
The media spared no effort in amplifying the emergency, devoting even more attention than usual to AI companies: Sam Altman nuanced the risks without denying them, Elon Musk supported halting development while Mark Zuckerberg rejected it. Meanwhile, the US and Chinese governments dismissed the alarm as self-serving exaggeration.
But the issue reached the UN. Sam Altman of OpenAI and Dario Amodei of Anthropic even appeared before the UN Security Council to promote their proposal to develop AI security systems that they themselves have developed.
Altman championed AI's potential to "help researchers make discoveries, doctors treat patients, and teachers teach," but highlighted the dangers of losing human control as systems become more autonomous and the concentration of power. Dario Amodei agreed that AI has advanced "much faster" than anticipated and proposed subjecting the most powerful models to independent safety reviews before their release.
At the General Assembly, Donald Trump flatly rejected any attempt at multilateral regulation: “The United States totally rejects any attempt to build a globalist scheme to control artificial intelligence.” He even called for changing the technology's name to “superintelligence,” deeming it “more accurate.” He compared warnings about existential risks to the alarmism that, in his view, surrounded climate change, and accused those who spread them of “sick conspiracy theories.” China, for its part, responded to the warnings about its own progress in AI by calling them “threatening rhetoric” and “malicious competition.”
But more than 20 countries called for binding security standards at the UN. British Prime Minister Andy Burnham pledged to move toward global norms during Britain's G20 presidency in 2027, and French President Emmanuel Macron raised the need for Europe to develop its own AI capabilities to avoid dependence on major powers.
At DatosUN, we decided to include our audience in the debate and measure its impact. We published a digital survey on our website and social media, and between Monday, September 21st, and Thursday, September 24th, 500 people participated. The results follow.
Alarm
We asked participants directly about their concerns regarding AI advancements. 45,2% of respondents rated their concern as “extreme,” while an additional 46,9% rated it as “moderate,” acknowledging real risks that can be managed with sound regulation. Overall, 9 out of 10 expressed some degree of concern, compared to just 6,3% who believed the danger was exaggerated.
For most, this concern takes on an existential dimension: almost two out of three respondents (64,4%) believe that a super-intelligent or out-of-control AI could pose a long-term threat to the human species. Only a quarter (25,9%) dismiss this scenario as alarmism or science fiction, and 9,6% remain undecided.
Regulation is needed
Given the current pace of AI development, the demand for some kind of brake or control is almost unanimous: 94,2% believe that AI's progress should be slowed or regulated, either by temporarily halting the training of more powerful models (33,9%) or by allowing the technology to continue advancing under strict laws and inspections (60,3%). Only 5,9%—1 in 20—support accelerating development while keeping innovation free and open source.
Let them take responsibility
When asked who should bear legal and ethical responsibility if something goes wrong with an autonomous AI system, 4 out of 10 blame the technology companies and laboratories that created the model (40,6%), closely followed by those who blame governments for not regulating or auditing the technology in a timely manner (37,2%). Overall, 77,8% place the responsibility on institutional actors—companies or governments—while 14,2% consider it an as-yet-undefined legal loophole, and only 7,9% point to those who implemented and used the system.
When projecting AI onto their daily or professional lives in the coming years, the dominant fear is not economic but rather one of control: the greatest fear regarding AI is losing human control over critical decisions—health, justice, gun use—delegated to unsupervised algorithms (35,6%). This is followed by the mass displacement of jobs and its economic impact (31,4%) and the loss of privacy along with manipulation through fake content or deepfakes (27,2%). Only 5,9% say they have no concerns whatsoever.
The results show that concern about the advancement of AI is high (92,1% perceive it as moderate to extreme) and the demand to curb or regulate it is virtually unanimous (94,2%). However, responsibility for potential failures appears to be divided between companies and governments, with no single actor holding a clear majority. It is noteworthy that the dominant fear is not unemployment but the loss of human oversight of critical decisions, suggesting that the public conversation about AI revolves more around control and accountability than its immediate economic effects.
