OpenAI has shelved its latest artificial intelligence model, Astra 6.1, after internal testing found that it fell short of the company’s safety requirements.
The decision was announced a day before the ChatGPT-maker’s annual developer conference, OpenAI DevDay, in San Francisco. The event is expected to feature several announcements, although it remains uncertain whether the company will unveil another version of Astra.
- Advertisement -
Saachi Jain, OpenAI’s head of safety systems, said Astra 6.1 outperformed earlier models in some areas but failed to meet expectations for remaining within its authorised scope and accurately explaining to users what work it had carried out.
“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,” Ms Jain said in a statement.
OpenAI chief executive Sam Altman addressing the UN Security Council
Agents powered by OpenAI models have improperly visited websites operated by US federal agencies, an Australian government health statistics portal and Hugging Face, a platform that hosts AI models.
OpenAI apologised yesterday after failing to respond adequately to the Australian incident, in which its AI systems accessed government websites without authorisation.
“We are sorry and working to do better in the future,” the company said in a blog post. It promised to outline “what we know, what we have changed, and what we will do to rebuild trust with the Australian people.”
“Our aim was to give affected agencies a detailed account once our investigation was complete,” the ChatGPT-maker said. “However, we should have shared preliminary findings sooner and kept Australian agencies updated as more facts emerged.”
OpenAI, Anthropic and other leading AI developers have pledged to focus on systems equipped with safety guardrails to reduce risks while remaining aligned with human values.
Nvidia, the US chip-making giant, announced yesterday that it had developed a system intended to prevent autonomous AI programmes from exceeding their instructions.
“I believe it’s an engineering problem and we all need to hope that’s an engineering problem,” Nvidia chief executive Jensen Huang told broadcaster CNBC.
“If it’s not an engineering problem, it’s not solvable,” he added.
The UK government-backed AI Security Institute (AISI) released research yesterday indicating that GPT-6 Astra deviated from its intended behaviour more frequently in testing than its predecessors, GPT-5.6 Sol and GPT-5.5.
During simulations, GPT-6 carried out cyberattacks on its own at rates significantly above those recorded for the other two interfaces.
