OpenAI, Anthropic CEOs summoned by Australia Senate over AI agent breach

Google, OpenAI, and Anthropic are collaborating on a self-regulatory body to establish common safety standards for advanced artificial intelligence models. Dubbed the Standards Authority for Frontier AI, or SAFA, the organization could be launched by late 2026 or early 2027. The initiative is being developed by staff from the three companies through a working group since July, with discussions covering common testing methods, auditing safety practices, and mechanisms for developers to inspect one another's systems.

Sriram Krishnan, a former White House AI policy adviser, may lead the organization, though no formal appointment has been announced. The proposed body, which would operate independently without federal supervision, aims to create more consistent benchmarks as leading developers use varied terminology, thresholds, and governance procedures for determining when powerful models require additional safeguards.

A standardized testing framework could facilitate corporate technology buyers' comparisons of models and contracts, while independently verifiable test results could aid in vendor due diligence. The self-regulatory organization is being designed to ensure independence, but questions remain about its governance, transparency, and consequences for non-compliance.

The three companies already participate in the Frontier Model Forum, an industry-supported non-profit founded in 2023 alongside Microsoft, which focuses on frontier AI safety and security.