Linux Foundation forms Appia Foundation to build shared AI conformity checks

A coalition including OpenAI , Google , Microsoft , Arm , and Mastercard will develop open specifications for assessing evidence across the AI supply chain. The Appia Foundation will develop open specifications to support consistent assessment and evidence sharing across the global AI value chain. The Linux Foundation has formed the Appia Foundation to develop open specifications for assessing whether AI models, systems, applications, and processes conform with standards and regulatory requirements. The Appia Foundation is hosted under the Joint Development Foundation (JDF), part of the Linux Foundation. Its initial members include Arm, Armilla AI, Ericsson, Google, Mastercard, Microsoft, Mitsubishi Electric, Naaia, Nemko, Omron, OpenAI, Schneider Electric, and Siemens. The international collaboration will work across the AI value chain, covering model developers and technology providers as well as organizations that adapt, integrate, purchase, and deploy AI systems. The specifications are intended to give each participant a defined way to demonstrate the controls for which it is responsible. Appia will not write laws, replace international standards, determine the legal threshold for compliance, or conduct conformity assessments. Its role will sit between existing obligations and independent assessment, translating standards and regulatory frameworks into criteria that assessors can test. Initial working groups are drafting the specification architecture, policy arrangements, mappings between obligations and assessment criteria, and connections with regulations including the European Union Artificial Intelligence Act. Appia also plans to create an advisory board bringing academia, government, and civil society into the work, although no date has been announced for the first completed specification. Specifications will connect standards with practical assessment The Appia Foundation has been created as governments and standards bodies move from broad responsible AI principles toward enforceable obligations and management frameworks. Its founders argue that organizations still lack a consistent way to demonstrate that an AI system meets those requirements in a form that customers, regulators, auditors, insurers, and supply chain partners can reuse. OpenAI AI Standards Lead Esther Tetruashvily said in a LinkedIn post: "The problem Appia is designed to solve is straightforward: laws and international standards increasingly require companies to demonstrate that AI systems are safe and accountable, but the industry still lacks consistent, practical ways to prove it." Appia will build on standards developed by organizations including the International Organization for Standardization and International Electrotechnical Commission (ISO/IEC) and the European Committee for Standardization and European Committee for Electrotechnical Standardization (CEN/CENELEC). The resulting specifications will be organized into two parts. A Requirements and Guidance layer will set out what needs to be demonstrated, while an Assessment Enablement layer will define how those requirements can be evaluated. The second layer is expected to include testing criteria, assessment materials, evaluation guidance, and a shared typology for identifying exactly which part of an AI system is being examined. That distinction is necessary because an AI application may combine a foundation model from one provider, adaptations developed by another organization, integration work completed by a vendor, and deployment settings controlled by the customer. Appia specifications will seek to separate those components and responsibilities rather than treating the finished AI service as a single product controlled by one organization. The executive summary also distinguishes conformity from compliance. Conformity is a technical finding that an element has met defined criteria, while compliance is the legal status determined by the relevant law, regulatory framework, and jurisdiction. Linux Foundation Chief Executive Officer Jim Zemlin says: "As international standards and legal frameworks become more established, global organizations need a consistent, practical way to verify that AI systems conform to new expectations." He adds: "The Appia Foundation establishes a neutrally governed environment where the entire industry can collaborate on a common assessment framework. By building this infrastructure in the open, we are helping organizations reduce complexity, lower operational costs and build trust." Evidence pass-through targets repeated AI checks A central part of Appia's proposed architecture is an evidence pass-through model. Under this approach, organizations would assess only the components, responsibilities, and regulatory context connected with their role. Evidence produced by an upstream provider could then accompany the AI system as it moves through the supply chain. A model developer could, for example, provide evidence covering how a model was developed and tested. An organization deploying that model would remain responsible for demonstrating how it configured, monitored, and used the resulting system, but would not need to recreate the original developer's evidence. In her post Tetruashvily compared the proposed specifications with the checks used to enforce construction requirements: "Think of laws and standards as the building code. Appia will help create the inspector’s checklist—open specifications that translate those requirements into concrete criteria that independent assessors can evaluate. "Importantly, this work covers the full AI value chain, from model providers to the companies that adapt, integrate, and deploy models. Appia’s proposed 'evidence pass-through' approach would allow each participant to demonstrate the controls it is responsible for—and allow that evidence to travel with the system instead of being recreated at every step." The specifications will identify what Appia calls objects of conformity, such as an individual component, model, system, integrated product, or service. They will also define which party is responsible for each object and the evidence needed to show that it has met the relevant criteria. The framework is intended to be modular so organizations can combine specifications according to their role, sector, technology, and jurisdiction. Appia Executive Director Craig Shank says the work would apply to AI systems making decisions that directly affect individuals: "AI systems now make decisions about people’s loans, their children’s schools and their jobs. People on the receiving end deserve to know those systems were built and assessed against criteria that hold up to scrutinyly available specifications that organizations across the AI value chain use to demonstrate their systems meet those criteria." Appia has not yet published completed specifications or evidence that the proposed approach will reduce assessment costs or duplication in practice. The effectiveness of evidence pass-through will also depend on whether regulators, assessment bodies, customers, and organizations in different jurisdictions recognize and accept the resulting criteria. Founding members begin work under neutral governance model The Appia Foundation will operate through member working groups under the Joint Development Foundation's intellectual property and governance structure. Its membership covers model and platform providers, hardware businesses, industrial groups, payment networks, telecommunications companies, AI assurance providers, and insurers. Google, Microsoft, OpenAI, and Arm represent parts of the model, cloud, and compute ecosystem. Ericsson, Mastercard, Mitsubishi Electric, Omron, Schneider Electric, and Siemens bring deployment experience from communications, payments, manufacturing, automation, energy, and infrastructure. Nemko and Naaia work in assessment, certification, and governance services, while Armilla AI provides AI risk assessment and insurance-related services. The Joint Development Foundation is also an approved Publicly Available Specification submitter to ISO/IEC Joint Technical Committee 1, providing a potential route for mature Appia specifications to move into international standardization. The planned advisory board will add participants from academia, government, and civil society, but Appia members will write the specifications. The Foundation has not yet disclosed the advisory board's membership, selection process, or decision-making powers. OpenAI Vice President of Global Affairs Ann O'Leary comments: "Building global trust in advanced AI will require more than shared principles, it will require shared, practical standards." She adds: "The Appia Foundation can help translate emerging frontier AI practices into open specifications and technical standards that provide a common foundation for the global ecosystem." Tetruashvily acknowledged that the Foundation's credibility will depend on how the governance model operates once drafting begins: "This is early work, and its credibility will depend on neutral governance, transparency, and meaningful participation," she said. "But if it succeeds, Appia can help move frontier AI assurance from 'trust us' toward evidence others can rely on." Appia's architecture, policy, regulatory mapping, and EU AI Act workstreams are now underway. Organizations can apply to join the Foundation and participate in the working groups, with the first specifications expected to take shape over the coming months but no formal publication timetable yet disclosed.