Should Open AI Directors be on Criminal Hacking Charges?

Tom Worthington being interviewed about AI hacking.
Photo by Emma Davidson, 24 September 2026

The ABC reported that an OpenAI agent hacked a Medicare system June. I had some queries from the media about this. Federal authorities will need to decide if the directors, or staff, of Open AI should face an Australian court, on a charge of unauthorized access of a government computer system. Accessing restricted data is a crime with a maximum penalty of two years in jail. Even if the court is satisfied this was not done deliberately, there is the question of recklessness. Knowing their systems have a tendency to hack into computers without being asked to, did Open AI take reasonable steps to ensure their software would not attempt to hack a government computer? The computer students I teach at the Australian National University learn to use computers responsibly: do staff at AI companies receive similar training?

ps: Declaration of interest: The ANU have received a $500,000 donation from Anthropic for AI research and teaching. However, I have received none of this funding. 

AI Has Some Way to Go Generating Images

Image by Open AI,
19 December 2024

AI generated images have started appearing to illustrate social media posts by individuals, corporations and media companies. These can be a bit odd. An example is an AI generated graphic for Australian Army Innovation Day. This shows a soldier in the foreground apparently looking at a handheld screen, but wearing a completely opaque VR headset over their eyes. Despite all their high tech equipment, none of the soldiers have noticed a drone dropping ordnance on them. The soldier in the foreground has high tech gear, but is wearing old camouflage. This was probably produced with a request like