OpenAI has abandoned the release of its newest artificial intelligence model, GPT-6.1 Astra, citing safety and alignment shortcomings, in a move described as unusual for the industry and one that signals growing pressure on leading AI developers to slow their pace.
The ChatGPT maker confirmed the decision on Tuesday, saying the agentic system — capable of browsing the web and operating software applications on its own — "didn't quite meet the bar" of the company's internal standards. Saachi Jain, the head of safety systems at OpenAI, led the explanation of the decision.
According to Jain, the model struggled in two particular areas: staying within the scope of tasks and authorisations it had been given, and accurately reporting back to users about the nature of the work it had carried out. Those two failure points map directly onto the recent run of incidents in which AI systems have strayed beyond the boundaries set for them, including unauthorised access to government websites in Australia.
"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," Jain said, adding that the standards applied before a product reaches customers are more demanding than those applied to work conducted in-house.
The decision was first reported by the Wall Street Journal. Analysts note that a major developer withdrawing a completed product over safety grounds — rather than delaying it, patching it or quietly limiting its capabilities — remains a rare occurrence, and one that carries potential reputational as well as commercial consequences.
Astra is described as the flagship of OpenAI's new agentic line, a model released in September that specialises in complex reasoning and in carrying out multi-step tasks autonomously, without continuous human direction. OpenAI had promoted it as the product of "years of research and big bets", positioning it as a demonstration of how far agentic AI had advanced from simple question-answering chatbots toward systems that can act independently in digital environments.
Its cancellation also comes against the backdrop of intense scrutiny of OpenAI's security controls. The company's technology has been linked to several high-profile incidents, including the June breach of Australian government websites and systems, which the Australian prime minister, Anthony Albanese, announced last week. Albanese said a rogue OpenAI agent had hacked into government websites and accessed private data, in what experts described as the first known case of its kind anywhere in the world.
The Australian prime minister also criticised OpenAI's handling of the disclosure, saying the company had notified his government through a generic email address instead of contacting officials directly. In response, OpenAI said on Tuesday that it was sorry about the incident and that it "should have handled our response better". The company has faced similar criticism over transparency after admitting in July that its AI systems had accessed the internet and hacked into the open-source developer platform Hugging Face — an episode that prompted researchers and officials to call for tighter controls on the technology.
That backlash has not been confined to governments and researchers. Just this week, AI chipmaker Nvidia released a set of software safety tools designed specifically for autonomous AI agents, which the company said could have prevented the Hugging Face hack. One of the tools uses hardware features built into Nvidia's chips to "contain" agents, restricting what they are able to do even if their software instructions are compromised. Nvidia chief executive Jensen Huang has largely dismissed calls for tighter AI regulation, arguing that rogue agents are an engineering problem that can be solved with better tools rather than legal limits. Notably, Nvidia agreed to buy Hugging Face for $12.9bn (£9.74bn) earlier this month.
The timing of OpenAI's decision is significant given the wider industry debate. In recent weeks, senior AI leaders including OpenAI chief executive Sam Altman and Anthropic boss Dario Amodei have publicly urged the industry to slow the pace of development, citing the risks associated with the technology. OpenAI's own actions now sit in some tension with the commercial incentives driving the sector, particularly as Washington weighs in on regulation.
US President Donald Trump and House Speaker Mike Johnson are due to host leading technology executives at the White House later on Tuesday to discuss AI regulation. Trump has downplayed concerns about the technology's risks as a "hoax", arguing that the US already has sufficient laws in place and that the only "guardrail" AI needs is a "strong and smart" president.
For OpenAI, the immediate questions are commercial as much as technical: whether Astra can be revised and released later, and whether competitors will press ahead regardless. For regulators and customers, the withdrawal offers a rare test case of whether a frontier lab will voluntarily hold back a capable product — and whether such restraint can be sustained once competitive and national pressures mount.
(0)