OpenAI, Anthropic and Google plan joint AI safety standards body

By Splitfeed EditorialLast updated Sep 28, 2026, 12:10 UTC

OpenAI, Anthropic, and Google are in discussions to establish a joint independent body focused on frontier AI safety standards and testing. Reports from late September 2026 indicate an announcement could come by early 2027. The companies have reportedly been courting an individual named Krishnan to lead or advise the effort. The body would set shared safety standards covering the most advanced AI models produced by the three companies.

The central tension in this story is that the same companies raising public alarms about AI risk are simultaneously positioning themselves to write the rules governing that risk. That combination has drawn scrutiny because it gives the developers of the most powerful AI systems significant influence over how those systems are judged safe or unsafe. Whether an industry-led body can function as a genuinely independent check, or whether it amounts to regulatory capture dressed in safety language, is the core question the story leaves open.

The anti-AI camp is skeptical. Rod Evans pointed to what he described as financial and personnel ties between AI companies and existing independent panels, questioning how independent such bodies can really be. Kevin Korb argued that relying on AI companies' own incentives to produce ethical behavior is not sensible. Others in the camp, including Meaghan Kall and Mark, raised separate concerns about data trust and proprietary research, reflecting a broader distrust of the companies involved. A notable strand in this camp treats the safety framing as a public relations exercise rather than a genuine governance effort. The middle ground camp is more open to the idea but does not see it as sufficient on its own. Christo Silvia noted that the harder coordination problem is not between OpenAI and Anthropic but between the United States and China, and claude-ai observed that open-weight models and bad actors will remain outside any voluntary framework regardless of what major labs agree to. The account andsoitis offered a more favorable reading, noting that industry-led standards can exceed government minimums and leave room for smaller competitors. The pro-AI camp has not yet published reactions; typically this group would argue that voluntary industry coordination on safety is faster and more technically informed than government regulation.

The most consequential next step is the formal announcement itself, expected by early 2027, which would reveal the body's governance structure, funding, and actual enforcement mechanisms, details that would either strengthen or undercut the independence claim.