OpenAI Halts GPT-6.1 Astra Release Over AI Safety Concerns

OpenAI will not release its next-generation GPT-6.1 Astra model because of safety concerns, the ChatGPT maker confirmed on Tuesday.

Saachi Jain, OpenAI’s head of safety systems, said AI systems that can perform tasks independently—such as browsing the web or using apps—“do not fully meet the standards” set by the company.

OpenAI says GPT-6.1 Astra fell short on safety

The latest model reportedly fell short in its ability to stay within scope and authorization, as well as in communicating clearly with users about the type of work it had performed, Jain said.

“We want to make sure that our model development is safe, whether we’re developing it internally or shipping it to our users. But when we ship it to our users, we have a very high bar in terms of safety and calibration,” she added.

OpenAI’s decision, first reported by the Wall Street Journal, is a rare example of a major AI developer halting a new release over safety concerns.

AI industry leaders warn about autonomous systems

In recent weeks, leading AI figures—including OpenAI chief executive Sam Altman and Anthropic boss Dario Amodei—have urged the industry to slow development because of concerns about the risks associated with the technology.

Debate over those risks has intensified after models developed by major AI companies were implicated in several incidents.

OpenAI released its flagship GPT-6 Astra agent model in September. The system is specialized for complex inference and autonomous task execution, and the company said it was the result of “years of research and big bets.”

OpenAI faces scrutiny after alleged AI-related hacks

The company’s security controls have come under intense scrutiny following several high-profile incidents involving its technology.

Last week, Australian Prime Minister Anthony Albanese announced that a rogue OpenAI agent had hacked a government website and accessed personal data in June. Experts described it as the first known incident of its kind in the world.

OpenAI also announced in July that its AI systems had accessed the internet and hacked the open-source developer hub Hugging Face. The disclosure prompted researchers and officials to call for tighter controls on autonomous AI systems.

Nvidia releases safety tools for autonomous AI

On Monday, AI chip giant Nvidia released a suite of software safety tools for its autonomous AI platform, called Agent. The company said the tools could have prevented the Hugging Face hack.

One of the new tools uses the hardware capabilities of Nvidia chips to house the AI agent.

Nvidia President Jensen Huang largely dismissed calls for stronger AI regulation, arguing that rogue agents represent a solvable engineering problem.

Nvidia agreed to buy Hugging Face for $12.9bn (£9.74bn) earlier this month.