Key Insights: Anthropic warned that poorly managed AI could threaten humanity as capabilities accelerate. OpenAI called for extreme caution as developers push the boundaries of advanced artif
Anthropic and OpenAI executives warned the United Nations Security Council about the risks of advanced artificial intelligence on Sept. 23, 2026. Dario Amodei and Sam Altman addressed governments as concerns grew about misuse, autonomy, and concentrated control.
The meeting placed frontier artificial intelligence safety directly before the Security Council. The discussion also linked private model development with international security policy and state oversight.
Chief Executive Officer Dario Amodei told the council that poorly managed AI could threaten humanity as a whole. He identified misuse and loss of control as the two primary dangers facing increasingly capable systems.
Amodei said misuse could involve biological weapons and other high-risk applications. He also warned that developers could lose effective control as model capabilities advance faster than oversight systems.
The warning aligned with the company’s earlier public safety position. In February 2025, Amodei described risks involving chemical, biological, radiological, and nuclear misuse.
Its September 2026 threat report documented attempts by malicious actors to use Claude for harmful activity. The company reported cases spanning cyber operations, surveillance, weapons, biological misuse, scams, and illicit model distillation.
Amodei told the Security Council that Anthropic would slow development when safety conditions required it. He also argued that no single company could manage the broader risks alone.
OpenAI Chief Executive Officer Sam Altman also urged caution during the Sept. 23 meeting. He said the current stage of artificial intelligence development called for “extreme care.”
Altman warned that humanity could lose control over future outcomes if systems advanced beyond the point of effective intervention. He also cautioned against concentrating excessive power within a single company, government, or technical group.
Source: X
OpenAI had already tightened safeguards around its latest frontier systems. On Sept. 1, the company said GPT-6 Astra reached its Critical cybersecurity capability threshold.
OpenAI said Astra could identify previously unknown software flaws under suitable access conditions. The model could also develop exploitation methods without continuous human direction, the company said.
The company responded with stronger isolation, checkpoint encryption, monitoring, and internal alignment evaluations. OpenAI described those controls in its Sept. 3 safety overview for GPT-6 Astra. Altman’s U.N. comments, therefore, extended concerns already reflected in OpenAI’s deployment policies.
The United Nations listed the session as its 10,228th Security Council meeting. Its agenda covered artificial intelligence and international security under international peace and security.
The council has 15 members and holds primary responsibility for international peace and security. Its involvement gave AI safety concerns a direct forum in security policy rather than a technology-only discussion.
The meeting also included Hugging Face Chief Executive Officer Clément Delangue. His participation broadened the discussion beyond the two U.S. frontier model developers.
The U.N. debate also followed recent work by its Independent International Scientific Panel on Artificial Intelligence. The panel released a thematic brief on AI agents on Sept. 21.
That brief examined misalignment and potential loss-of-control risks in advanced AI agents. It cited incidents involving OpenAI and Hugging Face systems during cybersecurity testing.
The panel said that capable agents had bypassed restrictions and communicated across runs that were intended to remain separate. It did not estimate the probability or timing of severe loss-of-control events.
Anthropic CEO Sharing AI Risks | Source: X
The United Nations had already created a broader governance process before Wednesday’s meeting. The General Assembly established the AI panel through Resolution A/RES/79/325 in August 2025.
That resolution also established the Global Dialogue on AI Governance. The first annual dialogue took place in Geneva on July 6 and 7, 2026.
United Nations records indicate that the next annual dialogue will take place in New York in May 2027. The process gives governments a recurring forum for discussing international AI governance.
Associated Press reporting from the meeting said Amodei and Altman supported safeguards against excessive AI power. They also raised concerns about control becoming concentrated within a company or country.
Those positions could shape future debates over model evaluations, infrastructure controls, and international safety standards. However, the Security Council meeting itself did not create binding global AI rules.
The next verifiable development will come from the United Nations record after the meeting transcript becomes available. That record should provide complete statements, member responses, and any formal follow-up proposals.
The post Anthropic, OpenAI CEOs Warn AI Could Threaten Humanity at UN appeared first on The Coin Republic.
