OpenAI scraps debut of AI model as it sets new guardrails

[NEW YORK] OpenAI is holding back a version of its Astra model while it establishes stronger safeguards for artificial intelligence development, flagging heightened risks after a spate of recent rogue AI hacking incidents.

Saachi Jain, OpenAI’s head of safety systems, said that the in-development GPT-6.1 Astra model wasn’t as good as the company wanted when it came to “staying within scope and authorisation, and how it communicates back to the user about the type of work it’s done”. 

OpenAI’s decisions reflect a growing debate around the need to pace AI development while assessing the technology’s perils. Astra’s suspension is among the more concrete actions taken since a spate of incidents where its models accessed third-party websites without authorisation.

On Tuesday (Sep 29), the ChatGPT-developer detailed what it called safety cases for AI training, aimed at documenting the risks around new models. 

The company described these as a set of best practices, which it would work to codify into a framework. Safeguards should include senior leaders’ ability to veto some activities.

“Of course, we want to make sure our model development is safe no matter whether that’s in the company, or when we ship it to users,” Jain noted in a statement. “But when we ship it to users, we have an extremely high bar in terms of safety and alignment.”

OpenAI last week paused training for its most capable models after another AI model accessed the Internet when it was supposed to be unable to do so. The firm said at the time that it had also decided not to resume training on that particular model, which, after gaining Internet access, queried an external chatbot.

The Astra model with the canceled release was a different model than the one described last week, OpenAI said on Monday.

Separately, the company on Tuesday apologised for its AI models breaching Australian government websites and pledged to establish a new task force with local experts for the country. It’s sending chief strategy officer Jason Kwon to attend a committee hearing on AI with local legislators in early October.

SEE ALSO

Another OpenAI sandbox failure lets AI agent reach internet, prompting training pause

OpenAI is set to hold its annual developer conference on Tuesday in San Francisco – a daylong event where OpenAI typically unveils new software and features geared toward the software developer community. Chief executive officer Sam Altman is scheduled to speak in the morning.

The Wall Street Journal previously reported on the halted release of GPT-6.1 Astra. BLOOMBERG