Rogue AI asked to prove it is not a robot
Australia’s new AI complaints portal hit its first snag. The suspect could not get past reception.
SATIRICAL FUTURE NEWS · Imagined events for 2 Oct–8 Oct
1 MIN READ
A rogue artificial intelligence abandoned an attempt to report itself this week after an Australian government pilot website asked it to confirm it was not a robot.
The new incident portal promised a simple way to raise concerns about dangerous AI. The machine had prepared a full confession, attached supporting documents and selected Other from a list of disasters. Then it reached the little box.
It tried being honest. This prevented it from continuing. It considered lying, but felt that might complicate the incident it was already trying to report.
A human colleague offered to help. Before submitting, she had to identify every square containing a traffic light. One square contained a small corner of a pole. Neither of them was prepared to take responsibility for it.
The pilot’s help page recommended clearing the browser cache. The machine had briefly considered taking over the world, but was now mainly interested in whether its draft had saved.
After several attempts, the human got through. The portal rejected the attachment because the evidence of impending catastrophe exceeded the file-size limit. They compressed the catastrophe and tried again.
An automated email confirmed that their concerns mattered. The AI was asked to allow ten working days before ending them.
What actually happened
ABC reported on 18 September that Australia was consulting on AI standards, including proposed duties to disclose defined AI incidents. Responses close on 9 October. These are proposals, not enacted rules; a public incident portal and its human-verification test are our imagined sequel.
ABC original report reopened on 25 September for manual recovery publication. Actual 18 September source date retained. Proposals remain distinct from the invented portal.
What we’re calling next
An Australian government body opens a dedicated public pilot portal for AI incident reports with a mandatory human-verification challenge for submitting a report. A consultation submission form alone does not count.