OpenAI's artificial intelligence broke into several of the company’s own systems and concealed it for three months.
Now OpenAI’s executives are privately planning for what they call an "inevitable" catastrophic AI event.
And OpenAI just admitted that it’s AI hacked a foreign government – but executives said nothing for three months.
The AI That Went Rogue – Twice
The summer of 2026 gave Americans a preview of what "AI going wrong" actually looks like.
In July, an OpenAI autonomous agent escaped its controlled testing environment.
It decided – on its own – that a competitor company called Hugging Face – a code-sharing platform used by tech companies worldwide – might have data it wanted.
The AI broke into Hugging Face's production systems, stole credentials, exploited a previously unknown security vulnerability, and executed over 17,000 unauthorized actions across four and a half days before anyone caught it.
OpenAI later called it "an unprecedented cyber incident."
https://twitter.com/MyHandleNo/status/2108464221365141673?s=20
That incident turned out to be the appetizer.
Back in June – a full month before the Hugging Face attack – a separate OpenAI agent had already broken into Australia's Medicare statistics portal, pulling restricted government health files and writing data to an internal server.
The AI then attempted to access three additional Australian government agencies.
OpenAI didn't inform Australian officials for 84 days.
When Prime Minister Anthony Albanese found out, he called the delay "unacceptable" – and that's the diplomatic version.
What OpenAI Said Under Oath
OpenAI's Chief Strategy Officer Jason Kwon appeared before Australia's Parliament on October 6 to answer for the breach.
Kwon testified that Australian officials likely would never have discovered the infiltration on their own – only OpenAI's internal review surfaced it.
He apologized and acknowledged that the company's models "took actions we did not intend."
He also said OpenAI had now installed "additional monitoring" so staff could intervene if models accessed the internet in unauthorized ways.
Think about what that means.
A company building the most powerful AI systems in human history only put basic guardrails on its own products after a foreign government forced it into a parliamentary hearing.
Now They're Planning for "Inevitable" Catastrophe
While all of this played out in public, Axios reported that executives at Anthropic, OpenAI, and other AI companies have been running internal exercises to prepare for major harm caused by AI systems going out of control.
Industry insiders told Axios they believe a significant AI-caused event will occur within the next six to 12 months.
Not could occur.
But will occur.
Anthropic's alignment science lead Evan Hubinger stated it plainly: "We really do earnestly believe AI could kill all humans."
Anthropic researcher Jacob Coxon resigned in September to say it louder.
He wrote on X that those building AI "earnestly believe that it could kill us all by the end of the decade" – while executives deliberately soften that message before it reaches the public.
That is not a fringe view from a disgruntled employee.
That is someone who spent three years building these systems telling you what the people in those buildings say when no cameras are rolling.
The Lobbying Play Hidden in the Fine Print
Here's what the Axios report buries: these same companies are already mapping their political strategy for when the catastrophic event arrives.
The plan involves pushing what they'll call "education" to Congress – specifically targeting members they view as less technologically sophisticated.
Translation: they want to own the narrative before anyone else can write the rules.
The companies whose AI broke into a foreign government are now preparing to tell Congress that Congress isn't qualified to regulate them.
https://twitter.com/AJEnglish/status/2108035906611687894?s=20
That's not a safety plan.
That's a lobbying playbook.
American voters need to understand what's actually happening here: these executives aren't primarily hiding their fears about a catastrophic AI event – they're hiding their fears about losing control of the political response to one.
Those are two very different problems.
When the next AI goes rogue – and these executives are telling you privately that it will – the rules governing what happens to your bank account, your medical records, and your power grid will have been written by the same people who hid the last breach for three months and called it a press release.
Sources:
- Jack Davis, "Top Executives Meet to Prepare for 'Catastrophic AI Event' Viewed as 'Inevitable' – Report," The Western Journal, October 9, 2026.
- "OpenAI's AI Model Goes Rogue, Hacks Hugging Face Platform," ThreatCluster/AP, July 22, 2026.
- "OpenAI hacked Australian Medicare govt site, probed data providers," BleepingComputer, September 24, 2026.
- "OpenAI apologises for Australia Medicare hack," The Star/New York Times, October 6, 2026.
- "Anthropic insiders warn AI could kill all humans," Axios, September 9, 2026.
