The dispute over artificial intelligence safety is best read as a widening divide between leading developers seeking common rules for judging powerful systems and a US administration that has rejected a multilateral regulatory approach. OpenAI, Anthropic and Hugging Face have called for “national and international” standards covering the measurement of an AI tool’s capabilities, assessment of related risks, safeguards and the degree of human oversight maintained over such systems.
That appeal has met resistance in Washington. A key technology adviser to President Donald Trump rejected the idea of any new form of AI regulation at a UN conference. Washington’s technology envoy also dismissed calls for a coordinated global framework governing artificial intelligence systems. The US government has therefore rejected the three companies’ appeals for global standards to evaluate AI risk.
The disagreement matters because common benchmarks could give governments and companies a shared basis for comparing the safety claims made about AI systems. Without agreement on how capabilities and risks should be measured, the industry’s calls for oversight rest on standards that remain to be defined.
Sam Altman, OpenAI’s chief executive, joined Dario Amodei of Anthropic and Clement Delangue of Hugging Face in advocating shared benchmarks for evaluating risk. Altman urged governments to act jointly for “the collective good in the face of powerful new technology”, while calling for national and international benchmarks.
“We need common standards so countries can compare evidence, verify compliance, and have a shared language and understanding what is happening,” Altman said. He has called for benchmarks that address the measurement of AI capabilities, the assessment of associated dangers, applicable safeguards and the level of human oversight that remains in place.
Delangue said: “The global community needs stronger standards for monitoring and incident disclosure.” Amodei has separately called for a global slowdown in AI development and urged regulations that “pace the frontier” to mitigate the worst possible outcomes.
On 15 September, OpenAI said the world’s biggest AI companies were working to create a standards body. The brief does not specify how such a body would operate, when it might be established or which companies beyond OpenAI, Anthropic and Hugging Face would take part.
The Trump administration’s position differs from several initiatives under President Biden. In 2022, then-President Biden tasked the White House with developing the Blueprint for an AI Bill of Rights. The blueprint sought consumer protection from AI harms and the preservation of civil rights, data privacy and worker protections.
In 2023, Biden’s administration issued an executive order using the Defense Production Act to require AI companies to disclose safety results, develop standards and engage the government in auditing AI models. In 2024, the Department of Homeland Security convened stakeholders, including Altman, on the AI Safety and Security Board to develop strategies for identifying and mitigating AI threats to critical infrastructure and national security.
The board’s report set out roles and responsibilities for deploying AI across water, energy and communications infrastructure. It also reiterated the importance of independent audits and risk assessments.
A proposal for AI regulation includes three recommendations: placing independent safety monitors in frontier labs to identify problems, creating common safety standards among AI companies in democratic countries, and ensuring that the US and other democratic countries work with authoritarian governments to coordinate safety efforts. It also says regulation should require meaningful community engagement throughout the design, development, evaluation and deployment of AI systems.
It remains unclear whether the Trump administration intends to pursue alternative domestic safety measures. The administration’s rejection also leaves open how its stance will affect international discussions on AI safety and whether shared standards sought by the companies can gain support without US backing.
Comments (0)