Google, OpenAI and Anthropic reportedly plan an industry-led AI standards body, tentatively named SAFA, for launch by late 2026 or early 2027. It would set frontier-AI safety standards and, under current plans, operate independently of government.
According to The Information, SAFA would support third-party pre-deployment model testing, set guidance for reporting safety and security incidents, and define how labs’ voluntary safety commitments work in practice. It would also set qualifications for independent auditors. Working-group members are debating whether SAFA should conduct testing itself alongside the Center for AI Standards and Innovation; some say that center lacks resources to thoroughly test frontier models.
The companies initially envisioned a federal role, but the proposed public-private partnership stalled during the Trump administration. The Information reported that the White House told the three firms to reach industry consensus before government involvement; some company representatives remain optimistic about a future government relationship. Sriram Krishnan and Arati Prabhakar were reportedly considered for CEO, and Condoleezza Rice and David Friedberg for chair.
