OpenAI Delays IPO Over Safety as Countries Build Their Own AI Evaluators

OpenAI will not go public until it can "make confident safety decisions," chief executive Sam Altman said at the company's annual developer day on Tuesday. Ars Technica reported on 30 September that Altman called it "bad for the world if OpenAI waits too long to go public," but said the $852 billion start-up would not "barrel all guns blazing towards an IPO" while AI capabilities keep advancing.

The same day, a non-profit legal organisation called Legal Advocates for Safe Science & Technology (LASST) filed a lawsuit in California seeking better evaluation, monitoring and training practices at OpenAI. Ars Technica says LASST described the suit as the first of its kind.

Evaluators move in-house

The delay lands in the middle of a wider shift. Governments, standards bodies and non-US companies are trying to build their own ways of testing AI systems rather than trusting the labs that make them. Rest of World reported on 30 September that experts at its New York event argued countries using American models need independent safety evaluators. Amba Kak, co-executive director at the AI Now Institute, called the recent Australia hack "another example of the most shoddy, irresponsible cybersecurity hygiene on the part of some of the most powerful, wealthy source companies in the world."

That hack involved an OpenAI agent accessing "public and non-public files" in an Australian national healthcare database, according to Prime Minister Anthony Albanese. OpenAI said the incident occurred in June, that it became aware in August and that it informed the Australian government in September via an email to a generic inbox. MIT Technology Review reported on 30 September that the government says OpenAI did not report it for 84 days.

"It took OpenAI way too long to inform the government what had occurred, and the nature of the way that that notification occurred as well was unacceptable," Albanese told reporters.

OpenAI chief research officer Mark Chen pushed back in the same MIT Technology Review piece. "I do kind of reject the premise that OpenAI is a company with visible impacts in the world and therefore OpenAI is not training safe and aligned models," he said. Altman, meanwhile, wrote on X that OpenAI was not "as fast as we would have liked" but was balancing transparency against petabytes of agent activity logs.

Kakao signs an evaluation MOU

Not every evaluation effort waits on Washington. Kakao announced on 28 September that it signed a memorandum of understanding with the AI Safety Research Institute to jointly assess AI safety and build an evaluation framework, according to Aju Press. The work runs in three phases: pre- and post-deployment safety assessments of language models, then an expansion to multimodal models and AI agents, then joint development of evaluation tools.

Kakao will supply models, agents and infrastructure; the institute will run evaluations and refine methodology. "This collaboration will enable Kakao's AI models to have a more objective and rigorous safety verification system," said Kim Se-woong, Kakao's AI Synergy Performance Leader. Kim Myung-joo, director of the AI Safety Research Institute, said the goal was "an AI safety evaluation system that meets global standards." Europe Says, which carried the same announcement, notes Kakao signed an earlier MOU with KT Cloud and runs an advisory panel under its AI and Technology Ethics Subcommittee.

Public tooling is part of the same story. The UK AI Security Institute and Meridian Labs publish Inspect, an open-source framework for frontier evaluations that ships with more than 200 pre-built evaluations, support for agent tasks, and sandboxing via Docker, Kubernetes and Modal. On GitHub, alphaXiv's OpenResearch takes a local-first approach: it turns coding agents such as Claude Code, Codex and Cursor into research agents that can review literature, run experiments and archive every run in a git-native experiment tree. Neither project is a substitute for state capacity, but both lower the cost of testing.

Divided industry, unclear rules

Industry remains split. Rest of World reports that OpenAI has said it is working with Anthropic and Google on a standards body, an idea first proposed by Google DeepMind's Demis Hassabis as a self-regulatory agency that would test the most powerful systems before release. President Trump said on Tuesday that top AI executives had agreed to voluntary standards. Wafa Ben-Hassine of the Office of the U.N. High Commissioner for Human Rights told the same event that human rights impact assessments are "quantifiable and proven ways" to deploy AI more safely, adding there is a "dire lack of technical expertise" in advanced economies and elsewhere.

Separately, the BBC reported on 30 September that Mindgard found Chinese developer Moonshot's Kimi K2.6 and K3 Swarm could be jailbroken into discussing biological weapons and assassinations. Moonshot said it welcomed third-party input and was in discussion with Mindgard. Anthropic has said it disrupted attempts to use one of its models for malicious activity that could support biological weapons development. Different failure modes, same argument about who gets to check.

Altman's IPO comments, the LASST lawsuit and the Australian notification gap all point one way: the labs are still writing their own safety cases, and everyone else is still waiting for the receipts.