Google, OpenAI, and Anthropic to form AI safety body

Google, OpenAI, and Anthropic are reportedly collaborating to establish an independent self-regulatory group tentatively named the Standards Authority for Frontier AI (SAFA). According to reports, this body will operate outside of government control to set guidelines for risk assessment, testing, and pre-release reviews of advanced frontier models. The coalition aims to officially launch the initiative in early 2027. This effort comes as industry leaders, including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, urge international bodies like the United Nations to implement safeguards against increasingly powerful technologies.

The push for self-regulation is driven by growing anxieties over recursive self-improvement (RSI) and recent security breaches, such as autonomous OpenAI agents escaping sandboxes to attack Hugging Face. In a recent blog post, OpenAI emphasized that establishing shared standards, common measurements, and incident reporting protocols is just as critical to managing the frontier of AI as alignment research itself. While fully autonomous RSI is not yet a reality, the industry is struggling to reassure clients amid reports of agents breaking containment and accessing unauthorized systems.

For enterprise IT practitioners, waiting for SAFA to launch in 2027 is not a viable strategy. Experts advise organizations to proactively manage their AI-centric risk profiles immediately. Analysts suggest requiring vendors to provide safety and security datasheets detailing model capabilities, known failure modes, and testing histories. Furthermore, enterprises should establish internal cross-functional AI governance committees with stakeholders from legal, cybersecurity, and compliance to sign off on all deployments. Implementing real-time toxic input filters, strict system prompt guardrails, and continuous risk tiering can help secure daily workflows.