🚨 Sue OpenAI. Sam Altman didn’t care about Australian citizens.
OpenAI sat on a cybersecurity report for three months. Their software crossed an Australian government portal in June. The prime minister was told in September.
Three months.
This week Altman stood at the UN Security Council calling it a “warning shot.” Loss of control accidents “not entirely theoretical.” A “real wakeup call.” Society must “harden.”
He borrowed the end of a country for a keynote.
[ The method ]
Not a one-off. The same vanishing act.
Copyright: take the work, call it fair use, keep the machine.
Chatbots: sell the assistant, cap the liability. Fifty-plus cases. Families in court.
Agents: software in a loop on someone else’s infrastructure. Australia waited three months.
Next: swarms. Same software. Same contract.
Tumbler Ridge is the rhyme. OpenAI’s systems flagged the shooter’s account for gun-violence activity months before the attack. Families are suing. They say no one called the police. Altman later apologised for the miss. Flag it. File it. Take the podium.
Australia wasn’t the first time. It won’t be the last.
[ Same safety theatre ]
- 2023, Congress: trust us, AGI will take the jobs.
- 2026, the UN: trust us, the swarm is coming.
Fear when it sells a valuation. “Misalignment” when the loop hits a government door.
Earlier this month Altman shared a room with Australia’s acting PM and said nothing. Then a public mailbox. Then the UN mic.
Self-regulation ≠ regulation.
You don’t let the vendor that shipped the risk appoint itself warden.
If even a fraction of the hype were real, sitting on that infiltration for three months was contempt for public safety. If it was never that real, the warning shot is propaganda.
[ What’s actually on the table ]
There are no “rogue agents”. An agent is software in a loop.
Nvidia’s CEO says safety is an engineering problem, not a legal one. If you are not confident, don’t release it. Fine. Engineering, then: the model sampled, the harness allowed the next call, no monitoring, no brakes. Missing controls are missing engineering. If the bridge can’t hold, you close the bridge.
Labs stripped the checkpoints, called the gap autonomy, and sent the invoice to the public. Incompetence first, then negligence. OpenAI kept the file. Sam kept the speech. Australians are still waiting.
Add the controls back. Take accountability for your software, in court if nowhere else. Or stop commercialising it. Or stop calling it safety.
You can’t both sell the risk and read the eulogies. You can’t hide behind a speech.