If you haven't already seen the news, last week Australia's Medicare became the first government body in the world (that we know about, at least) to have been compromised by an AI "agent", in this case one developed by OpenAI. And everything from the Australian government's response to the way much of the media is framing the intrusion has, once again, let us all down.
After it was revealed that OpenAI knew about the breach back in June, only told Australian Prime Minister Anthony Albanese last week and then did so via email (not even a phone call!), to a general account that was only checked once a day, do you know what Albanese's response was? He told Altman he was “disappointed”.
Motherfucker, that company's AI agent just hacked the "statistics reporting service portal" of a government body! The one that has all of our health records (which, thankfully, OpenAI did not obtain)! If a person had done this they'd be arrested. But when a tech giant's piece of code does it, all the tech giant gets is a sad face from the Prime Minister? And, worse, a bipartisan stance that a breach like this means we should build more data centres?!?!
It's OpenAI’s code! They made the thing, they should ultimately be responsible for it, and if that agent goes and does illegal shit then that company should be facing the full weight of criminal and legislative charges. The fact all the government has said on this front is that they are exploring the "legal situation" and would like some “guardrails” has me wondering what every lawyer and cybersecurity expert has been doing for the past three years.
we have control mechanisms for this. it's called The Law. these guys should be held personally liable for these harms. the suicides, the thefts, the intrusions into computer systems. they should be sued and prosecuted as appropriate. problem immediately solved
— dr. rich traditions, JD (@helldude.bsky.social) September 25, 2026 at 5:53 AM
[image or embed]
Reporting on the matter has gone about as well as it could possibly have gone for OpenAI. From the BBC to the ABC, blame for the attack has been placed at the feet of the "OpenAI agent", as though it's a conscious and culpable entity. Even OpenAI's own statement, which says "In the course of [a review], our models took actions we did not intend", tries to separate the company from the agents.
Motherfucker, they're not people, they're lumps of code that you trained and made. Even if the hack wasn't intentional, you're still responsible, and any coverage of these actions that tries to anthropomorphise AI is, consciously or not, playing right into the hands of companies like OpenAI and Anthropic, because it gives them an easy means of shirking liability. It's starting to feel like every newsroom around the world that hasn't made "We Need to Talk About How We Talk About 'AI'" standard training for their journalists is doing their audience a disservice.
i really can’t bring myself to care about AI anthropomorphism debates. “is AI thinking?” i don’t know. is a calculator calculating? questions for the philosophers. what’s important is that we put sam altman in prison.
— Peter (@notalawyer.bsky.social) September 26, 2026 at 2:04 AM
Making matters worse is the fact that, just days after the Australian news, it was reported that OpenAI's agents had been messing with US government websites as well. In its defence, the company said "None of the incidents were breaches...but were examples of its technology behaving in unexpected and concerning ways". Imagine trying that specific defence in court as a company making literally anything else. "Sorry, your Honour, our poison wasn't poisoning people, it was simply behaving in unexpected and concerning ways".
Maybe I've just been radicalised by a combination of reading the news and expecting some degree of accountability to be applied to some of the world's most important and dangerous people, but the distinction between what would have happened to a person doing all this vs what is happening to the company responsible for its code is sending me mad.
bad news babe a rogue openai agent must have hacked our refrigerator last night and drank 15 beers
— huck mason (@tylerhuckabee.bsky.social) September 28, 2026 at 11:15 AM
