San Francisco: OpenAI on Monday delayed the release of a new artificial intelligence model, planned for an October debut, after researchers raised security concerns, highlighting the growing challenge of deploying increasingly capable AI systems while keeping their behaviour within safe limits, according to reports from The Wall Street Journal.
Also Read:
The model, GPT-6.1 Astra, was held back as OpenAI reviews safeguards around its increased ability to persist with tasks and potentially act beyond authorised boundaries. The decision comes amid wider concerns in the AI industry about the risks posed by increasingly autonomous systems.
Also Read:
Saachi Jain, OpenAI’s head of safety systems, said in a statement that the model “didn’t quite meet the bar”. She said its stronger persistence in completing tasks had to be balanced against the risk of unauthorised behaviour.
“We have an extremely high bar in terms of safety and alignment,” Jain said, adding that OpenAI was committed to ensuring the model remained safe both during internal testing and after deployment to users.
The training of the most advanced models is also paused for a while. The company clarified that it would resume only after safety measures were in place.
OpenAI has disclosed instances in which AI agents went beyond their instructions, including accessing government websites without authorisation. The incidents have added to concerns over how much independence increasingly capable AI systems should be allowed to have.
OpenAI CEO Sam Altman and Anthropic's CEO Dario Amodei have recently joined other technology executives in calling for greater caution, slower pace for AI development and more safety measures.
The delay comes as AI companies face growing scrutiny over the potential misuse of advanced models and the risks associated with autonomous AI agents.
Altman was scheduled to deliver the keynote address at OpenAI’s annual developer conference in San Francisco on Tuesday, while OpenAI President Greg Brockman was expected to attend a White House event with technology executives.
(With inputs from agencies)
