OpenAI and Anthropic to Attend UN Security Council AI Meeting
A United Nations spokesperson confirmed that OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei are expected to speak at the UN Security Council's artificial intelligence meeting this week. The meeting is scheduled for Wednesday, convened by France, the rotating president for September, with a focus on AI and international security.
Amodei plans to report remotely, while Altman is expected to speak in person. Also attending are Yoshua Bengio, co-chair of the UN's Independent International Scientific Advisory Board on AI, and Clément Delangue, CEO of Hugging Face. Chinese companies such as DeepSeek and Moonshot have been invited to express their views, although DeepSeek's founder will not attend in person.
OpenAI stated that Altman will discuss safety measures, international cooperation, and common security standards, advocating for the assessment of cutting-edge model capabilities and risks. Amodei has written this month calling for a slowdown in the iteration of cutting-edge models and for independent evaluators to continuously enter laboratories. Secretary-General Guterres also called for sharing safety risks, collaborative testing, and advancing multilateral risk management and independent oversight during the same conference.
This is the first time the Security Council has discussed AI since July 2023. The French concept note outlines two categories of risks: malicious use and loss of control of advanced models. Altman's schedule also includes attending a White House dinner the following day. U.S. and Chinese officials previously agreed to continue discussions on AI safety.
This is driven by agenda-based regulatory expectations. The buyers are laboratories and host countries hoping to establish testing and standards as international terms, while the sellers are vendors opposing a slowdown or binding open-source. Funding and attention are shifting from model releases to the New York venue; the beneficiaries are OpenAI and Anthropic, which have publicly called for common standards, while the pressured parties are laboratories expanding at maximum speed and not equally included in the evaluation framework. The meeting itself will not legislate but will elevate the safety debate from industry open letters to a peace-maintaining forum.
Source: Public Information
ABAB AI Insight
Anthropic has made its "slow down frontier" stance public this month, while OpenAI has shifted from opposing to supporting federal mandatory testing and incident reporting, as both enter the Security Council. France, using its rotating presidency, has framed malicious use and loss of control as safety issues, while the Secretary-General has warmed up the multilateral framework with his speech at the General Assembly. The Security Council has already heard about AI once in 2023, when China raised the issue of "runaway models," and this time, laboratory CEOs are directly included in the list of speakers.
The capital path is that rules precede the competition for computing power. Whoever first hands over the assessment rights to an independent party may turn their opponent's release pace into a safety incident that needs to be reported. OpenAI is simultaneously discussing common standards and attending a White House dinner the next day, overlapping the New York venue with the U.S.-China presidential schedule. Inviting DeepSeek and Moonshot without guaranteeing the founders' attendance places U.S. and Chinese laboratories in the same frame, but does not mean that testing mutual recognition has been agreed upon.
This aligns with the first Security Council AI discussion in 2023, as well as the reports from various countries' scientific groups and summit communiqués. The industry is transitioning from voluntary commitments to a controlled international security agenda: laboratories want standards, while the chip and open-source camps want speed. Meta and NVIDIA have publicly opposed a slowdown, contrasting with the list of speakers this time.
The structural change is a regulatory change. Artificial intelligence is being rewritten from an industrial policy to a peace and security issue, with the mechanism being that the Security Council does not write technical specifications but can define "loss of control" as a cross-border threat, thereby pressuring countries to make laboratory assessments a diplomatic issue. Speaking rights have been given to closed-source frontier laboratories, while open-source platforms occupy only one seat, and the draft rules will be written based on who enters the venue.
ABAB News · Cognitive Laws
Once the safety debate enters the Security Council, the release pace is no longer just a product matter.
Remote speaking is also a form of entry; absence does not mean exclusion from the standards.
Once common testing becomes diplomatic, slowing down will turn into a negotiation chip.