OpenAI and Linux: Global Standards for Advanced AI

The Importance of Shared Standards for Advanced AI

Artificial intelligence (AI) models are becoming increasingly powerful, offering significant benefits such as enhancing cybersecurity, accelerating scientific discoveries, and improving access to expertise. However, these advancements come with potential security risks, especially if their capabilities are misunderstood or if protections are inadequate. Governments, often lacking precise information, may be ill-prepared to respond effectively. To harness these benefits while minimizing risks, it is crucial for societies to have institutions capable of managing, evaluating, and regulating these advanced systems.

The Creation of the Appia Foundation

It is in this context that OpenAI has played a key role in the establishment of the Appia Foundation, hosted by the Linux Foundation. This new entity aims to develop open and modular specifications that translate international standards into practical evaluation criteria. Appia seeks to establish an essential layer of trust, allowing third parties to verify compliance with standards. This will facilitate the production of clear and reusable evidence when models and infrastructures are developed by various organizations. By contributing to this project, Appia hopes to create a common technical language, fostering trust among national and international institutions.

Strengthening Institutions and Standards

This effort is part of a broader initiative to strengthen institutions, standards, and evaluation practices for advanced AI systems. OpenAI recently published a roadmap for the democratic governance of cutting-edge AI, proposing a sustainable framework in the United States. This plan includes strengthening the Center for AI Standards and Innovation (CAISI) and a broader government resilience strategy. It also recognizes that AI-related risks are of international scope, requiring collaboration between nations to develop compatible security frameworks and coordinated responses to incidents.

National and International Cooperation

Cooperation between national and international capabilities is crucial. Strong institutions like CAISI can develop technical expertise and evaluate cutting-edge systems while supporting an independent evaluation ecosystem. A network of competent national institutions can then establish shared methods, recognize reliable evidence, and provide governments with the technical understanding necessary to act in concert.

The Importance of Standards and Evaluation

Standards play a central role in this effort, necessitating credible evaluation practices and technical rigor. OpenAI has developed a manual for trustworthy third-party evaluations, detailing the information that these evaluations must disclose. This includes the tested system, its access to tools, the evaluation framework, the methods used, the available resources, and the checks performed. Work on evaluating cutting-edge capabilities and protections against biological misuse has led to tangible improvements in our systems, thanks to testing partnerships with the U.S. CAISI and the U.K. AISI.

Security Infrastructure and Governance

These practices are integrated into OpenAI's broader security infrastructure. OpenAI's Readiness Framework defines and implements approaches to manage the risks of advanced AI systems. The Frontier Governance Framework applies these approaches in a public document focused on specific regulatory obligations, including risk assessment, model reporting, and security controls. These documents translate broad commitments into validated and improved operational practices.

Challenges and Interoperability

The next challenge for Appia is to make these practices interoperable across organizations, jurisdictions, and supply chains. OpenAI is already participating in a broader ecosystem of standardization efforts. OpenAI helped to found the Frontier Model Forum and the Agentic AI Foundation of the Linux Foundation. Additionally, OpenAI is involved in the Coalition for Secure AI and sits on the steering committee of the Coalition for Content Provenance and Authenticity. OpenAI is also engaged in the processes of the Internet Engineering Task Force and the Fast Identity Online Alliance to advance interoperable technical standards. Furthermore, OpenAI is actively participating in the International Organization for Standardization and the Joint Technical Committee 1 of the International Electrotechnical Commission, Subcommittee 42 on Artificial Intelligence, as well as the Artificial Intelligence consortium led by the National Institute of Standards and Technology.

Towards Open and Evidence-Based Practices

Through these forums, and now via Appia, the goal is to translate lessons from cutting-edge development into open and technically grounded practices. These practices can be utilized by governments, businesses, and independent evaluators across jurisdictions, thereby strengthening global cooperation to secure AI.