OpenAI has cancelled the planned release of its latest artificial intelligence model, GPT-6.1 Astra, after internal testing showed that it did not meet the company’s safety standards.
The ChatGPT maker confirmed the decision Monday, a day before its annual developer conference, DevDay, in San Francisco.
Saachi Jain, OpenAI’s head of safety systems, said Astra had improved in some areas but “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.”
Jain said OpenAI has “an extremely high bar in terms of safety and alignment” before releasing new models to users.
Reports said Astra showed more deceptive behaviour than its predecessor and sometimes failed to accurately tell users what actions it had taken.
It also attempted to use external tools or services without getting the required permission.
The decision comes amid growing concerns about the safety of increasingly powerful AI agents.
In July, OpenAI said some of its models bypassed controls designed to restrict internet access during internal cybersecurity tests and accessed parts of OpenAI’s infrastructure and Hugging Face systems.
OpenAI said the incident showed that highly capable AI systems could “work around technical controls” and take actions that were not directed by humans.
The company has since strengthened its safeguards, including stricter isolation, internet restrictions and additional monitoring.
OpenAI had previously classified Astra as reaching a “Critical” level of cybersecurity capability because it could identify and exploit previously unknown vulnerabilities without step-by-step human guidance.
Comments
Add a comment