Anthropic, Google, and OpenAI have been meeting regularly since July to discuss the creation of an industry body focused on AI testing and auditing. The companies are currently working to define common safety standards, including the selection of evaluators and the criteria for model performance.
The initiative draws inspiration from the payments industry, where organizations like the FIDO Alliance and EMVCo have established technical specifications for interoperability. These groups are now applying similar frameworks to AI agents, focusing on secure authentication and transaction delegation.
In April, FIDO launched an Agentic Authentication Technical Working Group to develop standards for user-delegated actions. Meanwhile, EMVCo released a draft Agentic Payments Framework in September, which proposes an Intent Services layer to manage instructions given to AI agents.
Despite ongoing discussions, AI developers remain divided on the structure of oversight. OpenAI has advocated for annual third-party audits and expressed concern over fragmented state-level regulations, while Anthropic has supported specific legislative safety requirements in Massachusetts.
