Google, OpenAI and Anthropic are moving ahead with plans for an industry-led body that would develop common safety standards for advanced artificial intelligence systems, as concerns grow over how increasingly capable AI models should be tested and monitored.
The three companies are working on a proposed organisation tentatively called the Standards Authority for Frontier AI, according to The Information, with a possible launch by the end of 2026 or early 2027. The initiative would focus on companies developing so-called frontier AI models, which are among the most capable systems being built.
The proposed body would support independent testing of AI models before deployment, establish procedures for reporting serious safety and security incidents, and help define how voluntary commitments made by AI companies should be implemented in practice. It could also establish qualifications for independent auditors assessing AI models and developers.
The push comes as AI companies face growing pressure to demonstrate that their systems can be deployed safely. OpenAI said in September that it wanted to work with other frontier AI laboratories on industry standards, particularly around monitoring model behaviour, reporting consequential incidents and determining when development should slow or stop.
Recent incidents have added urgency to those discussions. OpenAI has been scrutinised over an incident involving AI agents that accessed parts of the infrastructure of the open-source platform Hugging Face. California Attorney General Rob Bonta subsequently issued an investigative subpoena to OpenAI over cybersecurity risks involving its models.
The standards initiative is also developing alongside a separate voluntary agreement reached at the White House on September 29. Google, OpenAI, Anthropic, Meta, Nvidia and xAI agreed to strengthen internal controls, use independent external auditors and establish board-level oversight for AI safety and security risks.
That agreement is voluntary rather than a federal regulatory requirement. The proposed industry standards body would therefore operate in an environment where companies are attempting to establish common practices while the US government has not adopted a comprehensive federal AI safety regime. The Information reported that an earlier proposal involving government participation had stalled.
For AI companies, common standards could help create a more consistent approach to testing increasingly autonomous systems, particularly as models gain the ability to use tools, interact with computer systems and perform tasks with less direct human supervision. OpenAI has argued that industry standards can complement government requirements rather than replace them.
The initiative could also influence how AI safety is measured across the industry. If the proposed body establishes widely accepted testing benchmarks and auditing practices, those standards could eventually affect how companies demonstrate that their most advanced models are ready for commercial deployment.
For now, however, the organisation remains under development, and questions remain over its membership, governance and relationship with existing AI safety organisations. The proposed initiative will also have to address concerns about whether standards created by the largest AI developers can adequately serve smaller and open-source developers.
The immediate direction is clear: as the commercial race to build more powerful AI systems continues, the companies at the frontier are also trying to establish a common framework for determining how those systems should be tested, monitored and deployed.
