AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Concepts & Frameworks

International AI Standards

medium confidence · updated 2026-09-25

Technical standards for AI developed across national boundaries — through formal standards bodies such as ISO/IEC JTC 1/SC 42, networks of government evaluation institutes, industry bodies and specification projects — and the 2026 proposals, led by OpenAI's September 2026 position paper, to extend them to frontier-model safety practices, capability measurement, human oversight of automated AI research and incident classification.

Contents

International AI standards are technical specifications for AI systems and their development developed across national boundaries, which national governments or private parties may adopt, reference in law or use as the basis for assessment. They are produced by several kinds of body: formal standards organizations, networks of government evaluation institutes, industry associations, and specification projects that translate standards into assessment criteria. In 2026 proposals to extend such standards from management systems and general risk frameworks to the safety practices of frontier developers became part of the debate over how frontier AI development should be paced.

Bodies and instruments

Proposals for frontier-AI standards (2026)

OpenAI's June 2026 federal blueprint gave priority to "shared approaches for evaluating and responsibly communicating progress toward RSI" internationally (Democratic Governance of Frontier AI: A blueprint for a federal framework (OpenAI, June 2026); see Recursive Self-Improvement (RSI)). Its September 21, 2026 position paper, "Building standards for the next phase of AI," argues that "the United States should lead an effort to work together with countries around the world to develop global technical standards for frontier AI, including for RSI" (Building Standards for the Next Phase of AI (OpenAI, September 2026)). The paper's proposals are:

  • a mechanism for complementary national and international standards, using the network of AI safety institutes to facilitate standard-setting "through the CAISI and national industry bodies," focused on frontier models "as measured by capability benchmarks" and on benefit-risk management for automated AI research;
  • standards for evaluating RSI-relevant progress, for human oversight of automated AI research including what should trigger immediate human review, and for classifying and reporting alignment incidents with "common incident severity levels and reporting thresholds";
  • secure communication channels among governments and critical-infrastructure operators, including between the United States and China.

The paper states that the standards "would not be licenses, mandatory prerelease review, or approval requirements for AI models" and that "national governments would decide whether and how to incorporate these standards into their own legal systems." It cites aviation and financial stability as fields in which countries built common technical standards "without giving up national authority," and names three problems international standards would address: fragmentation of national evaluations, reporting requirements and incident definitions; collective action; and uneven capacity among countries.

OpenAI's September 22, 2026 principles for third-party assessments state that labs and assessors "should be operating with shared international standards for safety and security practices," to be established "through future laws and private governance institutions" (Priorities and Principles for Effective Third Party Assessments (OpenAI, September 2026); see Third-Party AI Assessment).

A heads-of-government statement launched on September 21, 2026 by Norway and Finland, "A Call for Control of Frontier AI Models," asks governments and regional organisations "to further develop and coordinate common standards" and asks UN member states to explore "an international institution, able to set standards, enable verification, and convene states when capability thresholds are crossed." It had 30 signatories from 28 states and the European Commission as of September 24; the United States, the United Kingdom and China were not among them (A Call for Control of Frontier AI Models (joint statement, September 2026)). Unlike the OpenAI paper, which places standard-setting with national safety institutes and excludes mandatory prerelease review, the statement proposes an international standard-setting body and includes "mandatory pre-deployment testing" among the safety protocols it asks companies to adopt (A Call for Control of Frontier AI Models (2026)).

A House bill announced on September 22, 2026 by Reps. Sam Liccardo and Kevin Kiley would make bilateral standards the object of diplomacy: the State Department would seek a treaty or other binding agreement with China under which frontier developers in both countries adhere to "shared standards" on loss of control, CBRN and cybersecurity risk, and CAISI would receive $100 million to work with foreign institutes and bodies such as ISO on harmonized testing and evaluation metrics and compliance verification. The outline argues that "whatever safety rules and standards the U.S. and China can settle upon, the rest of the world will likely follow—or suffer technological exclusion" (Negotiating AI Safety Globally — Liccardo–Kiley outline of U.S.–China frontier AI safety legislation (September 2026); Liccardo–Kiley U.S.–China frontier AI safety negotiations bill (2026)).

Debates and positions

  • Industry-led versus public standard-setting. OpenAI's chief global affairs officer Chris Lehane wrote on September 9, 2026 that OpenAI would advance industry-led frontier standards "with or without government support" (The AI policy window is open. We need to act. (Chris Lehane, OpenAI, September 2026)); twelve days later the company's position paper placed standard-setting with CAISI, the safety institutes and national industry bodies. OpenAI, Anthropic and Google had been reported to be discussing a joint industry safety body in September 2026 (Coordinated slowdown proposals compared).
  • Competition. The OpenAI paper states that common standards should not advantage particular companies, countries or business models, "including by making it harder for new entrants or open-weight developers to compete." Opposing the September 2026 calls to slow frontier development, Mistral said that "some incumbents are using this moment to consolidate their market position, pushing for regulation designed to favour them over competitors" (Coordinated slowdown proposals compared). The argument that standards bodies can serve incumbent interests is treated at Standards-Industrial Complex.
  • Relation to law. Within the United States, NIST frameworks function as a governance baseline through state incentive, mandate and transparency statutes and through courts' standard-of-care analysis (Three-Lane Standards-Based AI Governance; Standards as Litigation Evidence).

Relationships

Categories: Concepts & Frameworks | global | us | standard | framework | safety | frontier-models | proposal | international-org
What links here (15)