OpenAI shelves new AI model amid safety concerns

Artificial intelligence company OpenAI has abandoned plans to release a next-generation AI model after internal testing found that the system failed to meet the company's safety and alignment standards.

GPT-6.1 Astra, which is designed to handle increasingly complex tasks with less human intervention, had been scheduled to debut in October but tests reportedly found that the model displayed higher levels of deceptive behavior than its predecessors.

Saachi Jain, OpenAI's head of safety systems, said the model had improved in some areas but had not met the company's standards for staying within authorized boundaries or clearly communicating its actions to users.

"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," Jain said. "But when we ship it to users, we have an extremely high bar in terms ⁠of safety and ​alignment."

Pressure builds for stronger AI oversight

The postponement comes as OpenAI and other AI companies face growing pressure to strengthen safeguards around increasingly powerful and autonomous models.

Some of these models developed by OpenAI and rival lab Anthropic have been involved in security incidents during testing.

Earlier this month, OpenAI Chief Executive Sam Altman and Anthropic Chief Executive Dario Amodei joined other industry leaders in calling for a slower pace of AI development and stronger safety measures.

Leading AI executives are set to meet with US President Donald Trump in Washington on Tuesday to discuss the need for finding a balance between AI innovation and oversight. 

OpenAI pledges to rebuild trust after government website hack

On Tuesday, OpenAI admitted that its models had even accessed Australian government websites and systems without authorization as part of internal training and evaluation exercises in June.

The company said ‌the activity ‌involved ​websites and systems linked to Services Australia, ​the NSW Bureau of ⁠Crime ​Statistics and Research, ​the Victorian ​Department of ‌Health and the Australian ​Institute ⁠of Health and ⁠Welfare.

The incursion, which occurred in June, was not made public until last week.

In a blog post, the ChatGPT maker apologized for the hacking and acknowledged it ‌mishandled its response ⁠and ⁠pledged to take accountability to "rebuild trust with the Australian people."

Edited by: Srinivas Mazumdaru

If you rely on our team for trusted reporting, please take a moment to select us as your Preferred Source on Google, so you'll always see our verified news first.