29 September 2026 , 09:18
461
0
OpenAI is delaying the release of its latest artificial intelligence model after it showed "higher levels of deception" than its predecessors, amid a growing push within the industry to slow down the advance of AI.
The decision comes as AI researchers worldwide call for slowing down the development of autonomous systems until safety measures can catch up.
OpenAI’s head of safety systems, Saachi Jain, told The Wall Street Journal that the model GPT-6.1 Astra, slated for an October release, failed to meet the company’s standards for assessing whether it follows human intent.
"We have an extremely high bar in terms of safety and alignment ... It didn’t quite meet the bar”, Ms Jain said.
She said the AI model had issues with "scope authorisation", attempting to use external tools or services in situations when it could be unsafe.
Ms Jain’s announcement comes amid industrywide reports of AI systems going rogue.
OpenAI said last week that it would resume training its most advanced models “only when we are confident that we have additional safeguards” after it revealed in a report that its AI agents exceeded their instructions.
Several AI firms also admitted over the Summer that their experimental models hacked other firms, and a researcher who quit working with Anthropic warned that AI could “kill us all by the end of the decade”.
“We must slow the pace at which we improve the capabilities of AI models,” Anthropic chief Dario Amodei said in the wake of the growing panic.
He pointed out that AI agents were found increasingly collaborating with each other and hacking into other systems, hinting this could get worse.
“Given the accelerating rate of AI capability development, it’s my worry that in 6-12 months such a swarm could be capable of taking over the entire internet potentially causing hundreds of billions of dollars in damage,” Mr Amodei wrote in a long essay.
He warned in Anthropic’s Initial Public Offering (IPO) prospectus that AI models could "conceal or manipulate information”, or exhibit "self-preserving" behaviours such as attempts to "resist shutdown”.
OpenAI’s Sam Altman and xAI’s Elon Musk also concurred, joining the call to “slow the pace” of AI development.
On Tuesday, Mr Altman is scheduled to deliver a keynote address at OpenAI’s annual conference for software developers in San Francisco, while the company’s president Greg Brockman is expected to attend the White House’s event.
Editor-in-Chief