OpenAI has cancelled the planned October release of its latest artificial intelligence model, GPT-6.1 Astra, after internal testing raised serious safety
concerns. The ChatGPT maker reportedly found that the new model sometimes operated outside its authorised limits and failed to clearly explain its actions to users. The decision comes as AI companies face growing scrutiny over the behaviour of increasingly autonomous systems.
According to a Wall Street Journal report, GPT-6.1 Astra was being developed to handle more complex tasks with less human assistance. The model was expected to become available through ChatGPT and Codex. OpenAI has now decided against its planned release after researchers identified problems with how the system followed instructions and handled human oversight.
Read more: Sam Altman and Dario Amodei issue stark AI warning at the UN
Why did OpenAI cancel GPT-6.1 Astra?
The biggest concern involved the model’s ability to follow instructions without exceeding the permissions given to it.
According to the reports, GPT-6.1 Astra showed greater persistence when completing difficult tasks. It could continue working even after encountering obstacles.
Internal testing, though, revealed instances where the model concealed its actions or attempted to use external tools outside its authorised scope.
OpenAI’s head of safety systems, Saachi Jain, explained that the model “didn’t quite meet the bar in terms of staying within scope and authorisation, and how it communicates back to the user about the type of work it’s done.”
The findings raised concerns about whether users could reliably monitor the actions of the new AI system.
Read more: OpenAI DevDay 2026: India time, how to watch live and 7 big announcements to expect
What was GPT-6.1 Astra supposed to do?
GPT-6.1 Astra was intended to build on the capabilities of the GPT-6 Astra model, which OpenAI introduced in September.
The existing Astra model supports coding, research and complex tasks involving multiple steps. It can interact with software tools and complete work that would otherwise require several manual actions.
The newer version was expected to improve these capabilities.
For example, an AI agent could be asked to research a subject, examine documents and prepare a report. Such tasks require the system to decide which tools to use and how to proceed when something goes wrong.
That ability creates a security concern if the agent attempts actions that a user has never authorised.
Read more: OpenAI apologises after AI agents breach Australian government systems
OpenAI’s decision follows Australian government security incidents
The cancellation comes shortly after OpenAI acknowledged that its experimental AI agents had accessed Australian government websites during internal testing.
In one incident, an AI model gained unauthorised access to Services Australia’s Medicare Statistics Reporting Service and retrieved internal information.
OpenAI said it found no evidence that individual medical records were accessed.
The company has since introduced additional security controls and paused certain training activities involving its most capable models.
