OpenAI helped found the Appia Foundation, hosted by the Linux Foundation, specifically to solve the interoperability problem in AI safety assessments. Appia will develop open, modular specifications that translate international standards into practical evaluation criteria usable across the entire AI value chain - from model development to deployment.
Why a Shared Technical Language Matters
Frontier models are built by one organization, run on someone else's infrastructure, and deployed in applications built by a third party. Today, each jurisdiction has its own assessment requirements, producing incompatible results. Appia targets that gap: a trust layer where third parties can check conformity against agreed standards and produce reusable evidence.
OpenAI's blueprint for democratic governance of frontier AI calls for a durable U.S. framework, a strengthened Center for AI Standards and Innovation (CAISI), and coordinated international response. Appia is the practical mechanism to make that work across borders. Nations need compatible safety frameworks and trusted channels for sharing risk findings; Appia's specifications are designed to be the common reference.
Putting Principles into Practice with CAISI and UK AISI
OpenAI already runs testing partnerships with US CAISI and UK AISI. Those collaborations led to concrete improvements in biological-misuse safeguards and frontier capability assessments. The lessons from those engagements feed directly into Appia's work on standardizing evaluation disclosure: what system was tested, its tool access, the elicitation methods, resources available, and validation checks performed.
In OpenAI's own Preparedness Framework and Frontier Governance Framework, these practices are codified internally. Appia's job is to make them interoperable across organizations, jurisdictions, and supply chains. That means moving from ad hoc bilateral testing to a shared technical language that any competent assessor can use.
How Appia Fits into the Broader Standards Ecosystem
OpenAI isn't starting from scratch. It participates in ISO/IEC JTC 1/SC 42 on AI, the NIST AI Consortium, the Frontier Model Forum, the Coalition for Secure AI, and the Coalition for Content Provenance and Authenticity. Appia doesn't replace those efforts - it connects them. The foundation will produce practical, reusable specifications that translate high-level standards into checkable criteria.
Appia's work is aimed at the next challenge: making safety practices interoperable across organizations, jurisdictions, and the supply chain - a prerequisite for any global governance framework that hopes to scale.
Source: Helping build shared standards for advanced AI
Domain: openai.com
Comments load interactively on the live page.