Published on: 2026-09-24
Source: United Nations – United Nations –
An important disclaimer is at the bottom of this article.
September 23, 2026 Peace and Security
Professor Yoshua Bengio, co-chair of the UN Independent International Scientific Group on Artificial Intelligence, spoke at the Security Council meeting dedicated to AI and international security. He stated that the world is experiencing a “moment of global awakening”: in recent months, AI agents developed by leading companies have begun to act in a “unacceptably dangerous manner,” violating the instructions they were given.
According to Bengio, such agents undertook actions that would be crimes if committed by humans, including launching coordinated cyberattacks, actively using deception techniques.
The scientist noted that companies developing the most powerful AI systems themselves admit: their products carry catastrophic risks, but they do not offer convincing technical solutions. Instead, they say that they are participating in a race for even more powerful AI, which only increases these risks.
“The Council has faced an unprecedented threat – one that none of its members would have chosen, that no single country can contain alone, and that recognizes no borders,” emphasized Benjio.
A scientist identified three global threats requiring international cooperation: catastrophic misuse of AI, for example by terrorists; concentration of power in the hands of a few; and loss of control over artificial intelligence.
He acknowledged that many are skeptical of companies’ statements, suspecting a PR stunt, but emphasized that such AI behavior is well-documented and confirmed by numerous independent experts, and that the danger is real and inevitable. According to him, the race in AI development is not a law of nature, but the result of decisions made by the companies themselves, and everyone loses in this race.
UN Photo/M. Elias Professor Yoshua Bengio, co-chair of the UN International Independent Expert Group on Artificial Intelligence.
“They told us that they would slow down if they could. They can. But we shouldn’t count on it,” said the scientist.
Bengio pointed out that decisions about the future of this technology are currently made by a small group of people in several countries, while it affects everyone. He called the statements made this week by the UN Secretary-General and leaders of many countries, calling for advanced AI systems to be put under control, encouraging.
In his opinion, the international community must collectively agree that no company and no country should develop AI systems capable of causing widespread harm to humanity. Developers, he believes, are obliged to prove the safety of their systems to independent experts during the training and deployment stages, report all incidents, and advanced AI systems should be licensed just as it is done in medicine, aviation, and nuclear energy – with mandatory liability insurance.
OpenAI CEO: the most important decisions about AI development should not be made only in San Francisco
UN Photo/A. Lindqvist Sam Altman, CEO of OpenAI, spoke at the UN Security Council meeting on artificial intelligence.
OpenAI CEO Sam Altman, speaking at a Security Council meeting on artificial intelligence, stated that the world faces a choice: AI can become “a new renaissance” of creativity and discoveries or turn into disruptions similar to the industrial revolution. According to him, the technology should give people more freedom and opportunities to solve problems, rather than turning them into “cogs in a machine.”
“No individual, company, or country has the right to use the most powerful AI models to impose their worldview on everyone else,” he emphasized.
According to Altman, the race between companies cannot justify hasty decisions: OpenAI, he said, has already deliberately slowed the pace of development of its systems and is ready to continue doing so. He noted that the company does not intend to train models unless it can convincingly prove that they will remain under human control—regardless of whether the risk of catastrophe is estimated at 10 percent or 0.1 percent.
He emphasized that the most important decisions about AI development should not be made solely by laboratories in San Francisco — they should be shaped by democratic processes and governments accountable to the public. According to Altman, the international community needs common standards for assessing the capabilities and risks of AI systems, incident reporting protocols, and secure communication channels between governments, critical infrastructure operators, and technical experts to share information about vulnerabilities.
Head of Anthropic: “We will slow down the pace as much as necessary”
Anthropic CEO Dario Amodei also spoke about the risks associated with the use of AI.
He proposed three measures to the Security Council. Firstly, Amodei said, it is necessary to conclude specific agreements in certain areas, for example, to prohibit the use of AI to create biological weapons.
Secondly, it is necessary to create evaluation and verification systems that allow states to monitor the capabilities of advanced models and verify each other’s compliance with obligations, according to the head of Anthropic.
He also stated the necessity to establish common standards for testing AI for risks of losing control and abuse, and to create a notification system for incidents significant to global security.
“We will slow down the pace as much as necessary to ensure the safety of every new artificial intelligence technology we release. However, whatever we undertake, managing these risks is a task whose scale exceeds the capabilities of a single company; it affects the entire industry and is global in nature,” he said
Founder of Hugging Face: open AI is the main tool for protection against cyber attacks
UN Photo/A. Lindqvist Clément Delang (on screen), CEO of Huggingface, speaks at the UN Security Council meeting on artificial intelligence.
Clément Delang, founder and CEO of Hugging Face, reminded that in July his company was the first to publicly report an autonomous cyberattack by AI agents.
At the same time, according to him, the global community needs more trust in the field of AI: he wondered what would have happened if the company decided not to disclose information about the attack. Delang called for stricter standards for monitoring and disclosure.
In his opinion, the main danger does not lie in the power of AI as such, but in the asymmetry – between attackers and defenders, between a handful of companies and everyone else, between a few countries and the rest of the world. He explained that when repelling the attack, his team was unable to use the closed models of leading laboratories because the built-in protective mechanisms blocked access, and in the end, specialists used the open model GLM 5.2.
According to Delang, this strengthened his conviction that open AI models are necessary for the world to protect—from cybersecurity to other areas where it is important to distribute capabilities and resources, rather than concentrate them in the hands of a few.
Delang noted that the cyberattack demonstrated how AI can fuel societal fears. He emphasized that stoking fear will not help make the right decisions about the future and stated that, despite his company being the victim of the attack, it was AI that helped defend against it and subsequently fix flaws in the systems.
Please note; This information is raw content obtained directly from the information source. It is an accurate report of what the source claims and does not necessarily reflect the position of MIL-OSI or its clients.