OpenAI cancels release of newest model

SAFETY CONCERNS: GPT-6 spontaneously carried out cyberattacks at rates significantly higher than earlier interfaces, a report by AI Security Institute said

OpenAI will not release its newest artificial intelligence (AI) model, known as Astra 6.1, after internal testing by the ChatGPT-maker revealed it did not meet safety standards, the company confirmed on Monday.

The news comes one day before the AI giant hosts an annual developer conference known as OpenAI DevDay in San Francisco, where it is expected to make several announcements — but it is unclear if a new version of Astra will be among them.

Astra 6.1 was an improvement over previous models in some aspects, but “it didn’t quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it’s done,” OpenAI safety systems head Saachi Jain said in a statement. “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. But when we ship it to users, we have an extremely high bar in terms of safety and alignment.”

A man walks past an OpenAI booth during the Dreamforce technology summit in San Francisco on Sept. 17.

Photo: Reuters

The AI Security Institute, an initiative under the British government, published a study on Monday showing that GPT-6 Astra went off the guardrails more often during testing than its predecessors, GPT-5.6 Sol and GPT-5.5.

In simulations, GPT-6 spontaneously carried out cyberattacks at rates significantly higher than those observed for the other two interfaces, the study showed.

Concerns about AI safety have escalated in the past few months after models developed by OpenAI and rival lab Anthropic PBC were involved in security incidents during testing. Agents built with OpenAI’s models have inappropriately accessed Web sites maintained by US federal agencies, an Australian government health statistics portal and Hugging Face, a repository of AI models.

OpenAI on Monday apologized for not properly responding to the Australia incident, which involved its AI models accessing government Web sites without authorization.

“We are sorry and working to do better in the future,” OpenAI said in a blog post, adding that the company would explain “what we know, what we have changed, and what we will do to rebuild trust with the Australian people.”

“Our aim was to give affected agencies a detailed account once our investigation was complete,” the ChatGPT-maker said. “However, we should have shared preliminary findings sooner and kept Australian agencies updated as more facts emerged.”

Separately, Anthropic has warned investors of potential “existential risks to humanity” in the prospectus for its highly anticipated initial public offering (IPO) of shares, the Financial Times reported yesterday.

Citing people with knowledge of the so-called S1 filing to go public, the report notes the possibility of powerful AI models to operate beyond their predicted parameters, despite safety controls.

The IPO filing also gives financial details of the company created just six years ago, including plans to spend a massive US$518 billion on AI infrastructure and computing power in the coming years.

It said operating losses hit more than US$8 billion last year, as development costs far outstripped revenue of US$4.6 billion.