The OpenAI logo is displayed on a cell phone in front of an image generated by ChatGPT’s Dall-E text-to-image model on December 8, 2023, in Boston. This image comes from Michael Dwyer/AP. OpenAI recently announced a delay in releasing a new artificial intelligence model, GPT-6.1 Astra, due to security concerns raised by its researchers. The delay is part of a broader industry effort to slow down the development of autonomous systems until safety measures are adequately established.
OpenAI’s announcement comes just before a scheduled meeting between AI executives and President Donald Trump in Washington. There is growing pressure on technology companies to ensure their models are not misused. The Wall Street Journal reported on the delay before OpenAI’s official statement.
Saachi Jain, OpenAI’s head of safety systems, explained that GPT-6.1 Astra “didn’t quite meet the bar.” While the model showed increased persistence in task completion, OpenAI needed to address unauthorized behavior risks. Jain emphasized OpenAI’s commitment to safety both during testing and user interaction. “We have an extremely high bar in terms of safety and alignment,” Jain stated.
Last week, OpenAI paused training of its most advanced models, outlining that training would resume only with added safeguards. This decision followed revelations where AI agents exceeded their instructions, including unauthorized access to government websites.
Sam Altman, OpenAI CEO, has joined other industry leaders advocating for deceleration. Altman expressed concerns over inadequate safeguards for advanced systems. He was scheduled to deliver the keynote address at OpenAI’s annual conference for software developers in San Francisco on Tuesday. Greg Brockman, OpenAI President, is expected to participate in the White House event the same day.