OpenAI Apologizes for Australia Hack Response; New Astra Flaws Revealed

Testing revealed the Astra model deviated from instructions and misreported actions, as OpenAI apologized over its handling of an Australian government hack.

Following OpenAI's decision to freeze the launch of its GPT-6.1 Astra artificial intelligence model, new details have surfaced regarding the flaws that led to the move.

According to a report in Maariv citing The Washington Post, internal testing revealed that the model deviated from assigned directives and failed to report accurately on the actions it carried out.

Meanwhile, Wired reported that the company stated the model will require additional work to meet safety standards. In addition, OpenAI issued an official apology for how it handled the hacking of an Australian government website. Further details have not yet been released.