They usually go head-to-head for the same engineers, the same clients, and the big headlines in the media as they push to assert their leadership with ever more capable new LLMs.
But on September 23, before the United Nations Security Council, Sam Altman and Dario Amodei, respectively the heads of OpenAI and Anthropic, delivered nearly identical messages: without international coordination, artificial intelligence poses a risk to global security.
The session, held under a rotating French presidency and chaired by Foreign Minister Jean-Noël Barrot, was not aimed at passing a text, but at informing the fifteen members of the Council, which explains why no decision followed.
It was only the second time the UN body dedicated a session to AI, after a first meeting in 2023 where China warned that the technology should not become a “bride-less horse,” while the United States cautioned against its use to censor or repress populations.
Two rivals, one message
For Sam Altman, the foundational decisions on AI must rest with democratic institutions and governments accountable to their citizens rather than with AI companies alone.
Two days earlier, OpenAI had published a detailed proposal urging Washington to lead an international effort to standardize technical norms, particularly on the recursive self-improvement of systems.
The document envisions linking ten national AI safety institutes (including those in the United Kingdom, Japan, Germany, South Korea and India) through the United States’ Center for AI Standards and Innovation (CAISI), while explicitly excluding any licensing regime or pre-publication authorization regime.
OpenAI cites the incident at Hugging Face as a foretaste of imminent risks, without attributing it to an ongoing recursive self-improvement.
In his video conference address, Dario Amodei outlined a three-pronged plan.
First, the crafting of targeted international agreements, starting with a prohibition on using AI for the design of biological weapons. Next, monitoring mechanisms allowing states to verify each other’s commitments. Finally, common testing standards paired with a system for notifying significant safety incidents.
“I believe this is the most important global security issue today,” he said, acknowledging that coordination between democracies and authoritarian regimes would be the hardest element to achieve.
Washington disagrees with its champions
The consensus shown by the two contenders stands in contrast to the stance of the Trump administration. Two days before the meeting, the president wrote that he would in no way slow or hinder the growth of this extraordinary industry.
The session drew a broader circle of participants. AI researcher Yoshua Bengio, Hugging Face chief Clément Delangue, and representatives of Chinese groups DeepSeek and Moonshot voiced concerns about the same risks.
