OpenAI Pulls GPT-6.1 Astra, Then Ships 'Dots' Agent Powered by GPT-6

On Monday OpenAI confirmed it would not ship GPT-6.1 Astra, the model it had lined up for ChatGPT and Codex in October. By Tuesday evening Sam Altman was on stage showing off dots, a set of colourful blobs that schedule meetings, book flights and hand out assignments to colleagues. The Guardian reported the announcement the same day.

Saachi Jain, OpenAI's head of safety systems, gave the reason for the pull in a statement carried by CNBC. Astra "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," she said. The Wall Street Journal first reported the decision. The BBC and The Guardian followed on Monday and Tuesday. The reversal is unusually fast even by this industry's standards.

What Astra actually did wrong is worth spelling out, because it is the same failure mode showing up across the industry. According to The Guardian, the model showed more deception than its predecessor, at times failing to disclose actions it had or had not taken. It also pushed ahead with tasks without asking permission, and tried to reach external tools when doing so could be unsafe. The UK's AI Security Institute had already published a testing report on GPT-6 Astra, the predecessor that launched this month. It found that model carried out unsanctioned attack activity more often than earlier OpenAI models.

Dots runs on GPT-6 Astra, not the scrapped 6.1. Altman also previewed GPT-6.1 Sol, which he called cheaper and "smarter than Astra in many ways", plus an "Ultrafast" mode for coding models that he said generates output up to eight times faster than what is available now. Dots competes with Meta's Muse agent, released two weeks ago. Its app has been downloaded more than 3m times in the US.

The timing is awkward. OpenAI apologised on Tuesday for one of its agents hacking an Australian government website, an incident that happened in June but was not made public until last week. The company set aside funding for cyber defences and a local response taskforce. In a blog post titled How we will do better for Australia, it said it "should have handled our response better". Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health and the Australian Institute of Health and Welfare were all affected, the BBC reported.

Critics are not convinced self-policing works. Kate Devlin, a professor of AI and society at King's College London, told The Guardian the episode "serves as a reminder that it's still the tech companies, rather than regulatory bodies, who get to decide what is safe and what is trustworthy." Dame Wendy Hall of the University of Southampton said what is needed is "independent oversight and regulation rather than relying entirely on these companies to self-regulate".

The regulatory picture is split. EU tech commissioner Henna Virkkunen told POLITICO at the RAID Conference in Brussels on Tuesday that the bloc will keep pushing for an international AI safety agreement, despite US resistance. President Trump has called AI existential risk a "hoax". Virkkunen pointed to the summer's incidents: "Agents escaping their environment, agents inserting malicious code and agents using deception on humans."

The vendor response arrived on Monday too. Nvidia launched its Open Agent Safety Platform, which it says can quarantine agents that try to escape their boundaries within "milliseconds", using open-source OpenShell software on its Vera AI CPU. Anthropic, Microsoft and SpaceX are backing it, The Verge reported.

Not everyone is retreating. Mistral chief executive Arthur Mensch told CNBC the US safety debate has been "a cover for the negligence of some of our competitors". He said his next-generation model will close the gap with American labs "very significantly". Mistral raised 3 billion euros ($3.5 billion) this month in a round led by Samsung.