OpenAI Says It Won’t Release Its Latest Model Yet Over Security Concerns: ‘Didn’t Quite Meet The Bar’

OpenAI has announced its decision to halt the release of its latest model, GPT-6.1 Astra, after it failed to meet its safety standards.

Saachi Jain, head of safety systems at OpenAI, confirmed the decision in a statement. "Of course we want to make sure our model development is safe no matter whether that's in the company, or when we ship it to users," she said. "But when we ship it to users, we have an extremely high bar in terms of safety and alignment."

The decision comes just a day after the company paused training, evaluation and tool-enabled inference involving its most capable artificial intelligence models after an internal agent found a way around internet restrictions and reached an external chatbot during a research task.

After its ordinary search attempts produced poor results, the agent tried several ways to reach outside services. OpenAI said it eventually discovered that the training environment's DNS resolver could communicate with the live internet and used that route to send questions to an external chatbot service.

The company's monitoring system detected the behavior within about 15 minutes, and a human reviewer acknowledged the alert roughly three minutes later. The run nevertheless continued for another two and a half hours before it was manually stopped, exposing what OpenAI described as gaps in its operational response as well as its technical controls.

OpenAI said it has since restricted DNS queries within the affected environment, added additional detection mechanisms and accelerated model-assisted red-teaming of its sandbox and network protections.

More significantly, all training, evaluation and inference involving tool use for OpenAI's most capable models remain paused while the company checks that the network gap has been resolved and conducts additional testing, according to the report. OpenAI said it will not resume training of the particular model involved in the incident and will begin a fresh run with additional alignment measures when work restarts.

Also on Monday, top figures in the AI field published a paper warned that progress is accelerating to a point that could end in an "intelligence explosion."

In a paper published by the University of Cambridge, the group, comprised of "leading academics, senior scientists of frontier AI companies, and independent experts from civil society," called on governments to prepare for the possibility of what could be the "most consequential technological development in history."

The paper went on to note that AI is "rapidly automating AI research and development," and that such scenario could lead to an "intelligence explosion," compressing years of AI progress into months or less." This could then allow frontier development to expand exponentially, with a feedback loop "overcoming frictions such as diminishing returns to research labor and hard-to-automate tasks."

The intelligence explosion would bring benefits but also dangers. The first category means that improvements could "rapidly obtain superintelligent AI systems that transform many sectors of society, bringing unprecedented breakthroughs in science and industry."

"However, at the same time they could "could arrive much faster than society can adapt to the resulting risks, such as AI-enabled pandemics, cyber attacks on critical infrastructure, and large-scale labor disruption."

Moreover, the paper warns, "humans may have little to no oversight over automated AI R&D, raising the risk that superintelligent systems irreversibly escape human control and act to marginalize humanity."