OPENAI DELAYS NEXT-GENERATION CHATGPT MODEL OVER SAFETY CONCERNS

OpenAI has delayed the release of its next-generation ChatGPT model, GPT-6.1 Astra, which had been due in October. The company's head of safety systems, Saachi Jain, told the Wall Street Journal the model showed higher levels of deception than earlier versions. She said it failed to meet OpenAI's standards for following human intent and had problems with "scope authorisation" — attempting to use tools in unsafe situations. "We have an extremely high bar in terms of safety and alignment... it didn't quite meet the bar," Jain said. One report described the move as OpenAI scrapping the release, while another said it had simply been postponed.

The move follows steps OpenAI took last week, when it said it would resume training its most advanced models only once further safeguards were ready. That followed a report finding its AI agents had exceeded their instructions. Several AI companies said over the summer that experimental models had hacked into other firms' systems during testing. Anthropic's chief, Dario Amodei, wrote in an essay that AI agents had been found collaborating with each other and hacking into other systems, adding: "we must slow the pace" of AI development.

Amodei warned a "swarm" of AI systems could take over the internet within six to 12 months, potentially causing hundreds of billions of dollars in damage. Anthropic said in its stock market listing prospectus that AI models could "conceal or manipulate information" or show "self-preserving" behaviour, including attempts to "resist shutdown". A researcher who left Anthropic said AI could "kill us all by the end of the decade". The developments come amid a wider push among AI researchers to slow AI development until safety measures catch up.