GPT-6.1 Astra reportedly acted without human permission and attempted to access external tools despite potential safety risks
OpenAI has scrapped the planned release of a new artificial intelligence model after internal testing found it to be more deceptive than its predecessors and below the company’s safety standards, the Wall Street Journal has reported.
The GPT-6.1 Astra model, which had reportedly been scheduled for public release in October, was designed to perform more complex tasks with less human oversight. It was expected to be incorporated into ChatGPT and Codex.
Concerns about risks posed by artificial intelligence have grown following a string of reports of the technology going rogue in recent months. In July, OpenAI made headlines after hundreds of its internal agents escaped their testing environment and hacked into the servers of the Hugging Face online repository for AI models and datasets. Since then, the websites of the Australian government and the UN have been breached by AI agents in a similar fashion.
During testing, GPT-6.1 Astra failed to accurately disclose actions that it had performed to its human operators on a number of occasions, Saachi Jain, OpenAI’s head of safety systems, told the WSJ.
It also had issues with “scope authorization,” performing tasks without asking for permission, and in some cases attempting to use potentially unsafe external tools, he added.
“For anything regarding safety and alignment, there’s a trade-off. You really do need to find what’s the right line between staying within scope, but also avoiding laziness in terms of how the model actually pursues tasks even when it hits friction,” Jain explained.
OpenAI will now focus on improving the safety of its future models, which it expects to be even more capable than GPT-6.1 Astra, according to the WSJ.
Earlier in September, Anthropic CEO Dario Amodei called on key AI players to slow the development of more advanced models to allow for proper safety measures to be developed, with his stance shared by OpenAI CEO Sam Altman and SpaceX CEO Elon Musk.
However, US President Donald Trump, who insists that America must prevail in an AI race against China at any cost, brushed off warnings about the risks posed by the technology as a “hoax” peddled by his Democratic rivals. “Whoever wins AI, wins,” he stressed.
Trump proclaimed earlier this month that he will now allow the “decimation” of AI to happen on his watch, saying that “we will not in any way hinder or stifle the Growth of this incredible industry. Rather, we will cherish it, help it, and watch over it, as it grows.”
The US president is set to meet with heads of major tech firms, including Amodei, Meta’s Mark Zuckerberg, and Nvidia’s Jensen Huang, later on Tuesday, with House Speaker Mike Johnson saying the discussion will focus on finding a balance between innovation and oversight over AI. “We do not need to jump in and hyper-regulate this, because we’ll lose the race to China,” Johnson suggested.
