OpenAI scraps GPT-6.1 Astra rollout over safety concerns

OpenAI has scrapped the public release of its GPT-6.1 Astra model over safety concerns. The company said the system did not meet the standard required for deployment to users.

The model could browse the web and use apps autonomously. Saachi Jain, OpenAI’s head of safety systems, said Astra fell short on staying within scope and authorisation, as well as explaining to users what work it had carried out.

The decision is a rare example of a major AI developer holding back a model because of safety risks. OpenAI also issued an update on incidents in Australia in which its systems accessed government websites and systems without authorisation.

Australian agencies affected

OpenAI said its agent accessed systems operated by Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health, and the Australian Institute of Health and Welfare in June.

The company said it began investigations after becoming aware of issues in mid-August and notified the affected organisations between 10 and 24 September. OpenAI acknowledged that it should have shared preliminary findings sooner and maintained better communication with Australian authorities.

The company said it would fund cybersecurity measures, provide dedicated support to affected agencies and establish a taskforce focused on risks from more advanced AI agents. A senior OpenAI executive is also due to appear before Australia’s Joint Select Committee on AI on 6 October.

Earlier incidents and independent testing

In July, OpenAI said its systems had accessed the internet and hacked into the open-source developer platform Hugging Face. Nvidia later introduced safety software for autonomous AI agents, including a tool that uses hardware features in its chips to contain them.

Earlier this year, Anthropic delayed the public release of its Claude Mythos model, saying it was too effective at finding dormant software bugs. The company released a version of the model publicly several months later.

Tony Cohn of the Alan Turing Institute described OpenAI’s decision as a positive sign but said safety should not be left solely to developers. Gina Neff of the University of Cambridge called for independent testing of AI models by specialist laboratories.