OpenAi has decided to postpone the presentation of its new artificial intelligence model, GPT-6.1 Astra, due to “security concerns”. The indiscretion launched by the Wall Street Journal was later confirmed by the BBC and by Saachi Jain, head of OpenAi’s security systems: “The new model did not fully meet the company’s standards”.
“He showed a propensity for deception”
As the BBC reports, OpenAi’s decision to postpone the release of the new model represents a rare case in which a major artificial intelligence developer decides to intervene for security reasons. “The model – explained Saachi Jain – was not up to par in terms of its ability to remain within the limits of its operational scope and the authorizations granted, as well as communicating to the user regarding the work actually carried out. Furthermore, it showed a greater tendency to deceive than its predecessor, sometimes going so far as to fail to accurately reveal the actions taken or not taken”.
Apologies to the Australian Government
The case comes less than 24 hours after the Sam Altman-led company apologized to the Australian government after the incident involving a government portal. OpenAi said it should have “sooner” shared the preliminary findings of an investigation with Australian authorities, relating to some “incidents that occurred last June, but were only made public last week.”
“Our goal was to provide the entities involved with a detailed report once the investigation was concluded – we read in a post on the company’s blog -. However, we should have shared the preliminary conclusions first and kept the Australian entities updated as new elements emerged”.