Google, OpenAI, Anthropic Plan AI Safety Standards Group

Google, OpenAI, and Anthropic are independently pursuing the creation of an AI safety standards organization, aiming to launch the self-regulatory body by year-end or early next year without government oversight. The organization, tentatively named Standards Authority for Frontier AI (SAFA), seeks to establish industry-led guidelines for advanced AI model development.

SAFA's planned core functions include supporting third-party testing of AI models before deployment and defining protocols for AI developers to report safety and security incidents. The initiative aims to translate voluntary safety commitments made by various AI labs into concrete practice standards.

The group is also discussing whether SAFA should directly oversee the safety and capability testing of AI models. Some members have expressed concerns that current testing resources, including those at the U.S. Department of Commerce's AI Safety Institute, may be insufficient for adequately testing frontier models.

Discussions are ongoing regarding how SAFA could supplement existing testing efforts without creating functional overlap. The companies involved have also publicly advocated for unified safety standards and increased international coordination in the AI field.