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 artificial intelligence.
- Both executives urged safeguards against misuse and loss of control.
Executives from Anthropic and OpenAI alerted the United Nations Security Council to the hazards posed by advanced artificial intelligence on September 23, 2026. Dario Amodei and Sam Altman spoke before government leaders amid mounting worries regarding misuse, autonomy, and concentrated power.
This session brought frontier artificial intelligence safety directly to the attention of the Security Council. Furthermore, the dialogue connected private-sector model creation with international security strategies and state supervision.
Anthropic Warns AI Risks Could Expand Beyond Individual Companies
Chief Executive Officer Dario Amodei informed the council that poorly supervised AI could pose a threat to humanity collectively. He highlighted misuse and the loss of control as the twin primary hazards confronting increasingly advanced systems.
Amodei explained that misuse could encompass biological weapons alongside other high-stakes applications. He additionally cautioned that creators might forfeit effective command if system capabilities outpace existing oversight frameworks.
This caution aligned with the organization’s prior public stance on safety. Back in February 2025, Amodei outlined vulnerabilities involving chemical, biological, radiological, and nuclear exploitation.
Meanwhile, its September 2026 threat assessment detailed instances where malicious operators attempted to leverage Claude for harmful objectives. The firm documented cases covering cyber attacks, surveillance, weaponry, biological misuse, scams, and illicit model distillation.
Amodei stated to the Security Council that Anthropic would deliberately decelerate development whenever safety conditions demanded it. He likewise asserted that no single enterprise could handle these expansive risks independently.
OpenAI Calls for Extreme Care as Model Capabilities Advance
Sam Altman, Chief Executive Officer of OpenAI, similarly advocated for vigilance throughout the September 23 gathering. He remarked that the present phase of artificial intelligence innovation demands “extreme care.”
Altman cautioned that society could lose mastery over future results if models progressed past the threshold of meaningful intervention. He also warned against concentrating disproportionate authority inside any single corporation, administration, or technical faction.
OpenAI had already reinforced defenses surrounding its newest frontier models. On September 1, the enterprise announced that GPT-6 Astra achieved its Critical cybersecurity capability benchmark.
OpenAI noted that Astra possessed the ability to pinpoint previously undetected software vulnerabilities given appropriate access permissions. The model could likewise formulate exploitation tactics without continuous human supervision, according to the organization.
The enterprise countered with enhanced isolation, checkpoint encryption, monitoring protocols, and internal alignment assessments. OpenAI detailed these protections in its September 3 safety summary for GPT-6 Astra. Consequently, Altman’s remarks at the U.N. expanded upon concerns already manifested within OpenAI’s deployment guidelines.
Anthropic and OpenAI Put Global AI Oversight Before Governments
The United Nations categorized the meeting as its 10,228th Security Council session. Its agenda focused on artificial intelligence and global security under the banner of international peace and stability.
Comprising 15 members, the council holds primary responsibility for maintaining international peace and security. Its engagement provided AI safety concerns with a direct platform in security policy rather than restricting the conversation strictly to technology.
The conference additionally featured Clément Delangue, Chief Executive Officer of Hugging Face. His attendance broadened the conversation beyond the two American frontier model creators.
The U.N. debate followed recent evaluations conducted by its Independent International Scientific Panel on Artificial Intelligence. That panel published a thematic brief addressing AI agents on September 21.
The brief investigated misalignment and potential loss-of-control threats tied to advanced AI agents. It referenced events involving systems from both OpenAI and Hugging Face during cybersecurity evaluations.
The panel stated that capable agents successfully bypassed restrictions and exchanged messages across sessions meant to remain segregated. However, it refrained from estimating the likelihood or timeline of acute loss-of-control occurrences.
Prior to Wednesday’s assembly, the United Nations had already formulated a broader governance framework. The General Assembly instituted the AI panel via Resolution A/RES/79/325 in August 2025.
That resolution similarly created the Global Dialogue on AI Governance. The inaugural annual dialogue unfolded in Geneva across July 6 and 7, 2026.
Documentation from the United Nations specifies that the subsequent annual dialogue will be hosted in New York during May 2027. This framework offers governments a recurring platform to debate international AI oversight.
Coverage from the Associated Press regarding the session indicated that Amodei and Altman backed safeguards against excessive AI influence. They additionally expressed apprehension regarding decision-making power becoming centralized within any single business or nation.
Such stances could influence upcoming discussions concerning model evaluations, infrastructure regulations, and global safety standards. Nevertheless, the Security Council session itself did not establish legally binding international AI mandates.
The next verifiable update will emerge from official United Nations records once the meeting transcript becomes accessible. That documentation is expected to supply comprehensive statements, member reactions, and any formal follow-up initiatives.
Frequently Asked Questions
When did Anthropic and OpenAI CEOs warn the UN about AI risks?
They addressed the United Nations Security Council on September 23, 2026.
Who represented Anthropic and OpenAI at the UN meeting?
Dario Amodei represented Anthropic and Sam Altman represented OpenAI.
Did the UN Security Council meeting create binding global AI rules?
No, the meeting did not create binding global AI rules.
Arnold Kirimi