OpenAI halts the launch of the latest version of ChatGPT for "security problems"

BarcelonaOpenAI will not launch, for the moment, its new artificial intelligence model, GPT-6.1 Astra, for security reasons, as confirmed this Tuesday by the company that created ChatGPT. The model—which can browse the internet and use applications autonomously—"did not quite meet the level required" by the company's standards, as reported by The Wall Street Journal.

OpenAI also reported this Tuesday on several incidents that occurred in June, but which did not become known until last week, in which its models accessed Australian government websites and systems without authorization. These incidents and other similar breaches have intensified in recent weeks the debate over the risks of artificial intelligence, in many cases promoted by the same large technology companies that develop it. Some of the sector's main executives, among them Sam Altman, from OpenAI, and Dario Amodei, director of Anthropic—responsible for the popular Claude model—, have called on the industry in recent weeks to slow down the pace of development due to concerns about the associated risks.

OpenAI's decision is a rare case. According to company sources, the model did not meet the standards regarding the scope and established authorizations, nor regarding the information provided to the user about how it performed the tasks. The flagship model GPT-6 Astra, which was launched in September, specializes in complex reasoning and the autonomous execution of tasks. OpenAI assured that it was the result of "years of research and big bets." The version that has been halted is 6.1.

OpenAI plans to hold its annual developer conference this Tuesday in San Francisco, and several announcements are expected. It is unclear whether there will be new information regarding the problems with the latest version of ChatGPT.

Hacking in Australia

Before the announcement, it had already been noted that AI exceeds limits in the automation of information search tasks. The Australian government reported last week that an OpenAI agent had accessed information from a government system related to health statistics and medical insurance in June. The Australian Prime Minister, Anthony Albanese, described the episode as "unacceptable." OpenAI alleged that it discovered the incident in August during a review of its models' activity. Albanese criticized the company for notifying him of the incident through a generic email address instead of contacting the responsible officials directly. OpenAI has regretted the events, admitted that it "should have managed the response better," and detailed that the affected departments included Health, Welfare, and the New South Wales Department of Interior.