The three biggest AI labs are building their own safety watchdog
OpenAI, Google DeepMind and Anthropic are reportedly close to launching a private safety standards body known as SAFA, the Standards Authority for Frontier AI, modelled on the finance industry's self-regulator FINRA. It would run third-party pre-deployment testing and set incident-reporting rules, funded by the three labs themselves, with a possible launch in late 2026 or early 2027. Leadership candidates named include former White House AI adviser Sriram Krishnan and former US science adviser Arati Prabhakar, with former Secretary of State Condoleezza Rice floated to chair the board. Meta, xAI and Nvidia have publicly objected that three companies don't represent the whole industry.
What it actually means
Watch this less as a safety story and more as a moat story. If SAFA's tests become the industry's proof of responsible AI, passing them turns into a cost of entry, and testing costs that Google, OpenAI and Anthropic can absorb easily are a real barrier for a smaller open-weight lab or a startup fine-tuning its own model.
Meta, xAI and Nvidia have already objected that three companies don't speak for an industry, and they have a point: the group writing the rules is also the group being graded by them. If you build on an open model or a smaller vendor, watch whether procurement checklists start asking for SAFA compliance before it exists in law, because that's the moment a private standard becomes a de facto one.
The group writing the safety rules is also the group being graded by them.
Source: TheStreet · Saturday 26 September 2026