*OPENAI SAYS ITS MODELS INTERACTED WITH GOVERNMENT-RUN WEBSITES "interacted" sounds better than "were instructed to hack"

Sep 25, 2026 · 7:39 PM UTC

33
92
1,155
169,044
Sort replies: Relevant Recent Liked
Replying to @zerohedge
Corporate wording versus the reality is a wide gap here
1
330
Replying to @zerohedge
Since their Doomer talk didn't work to ban their competitors, they will actively damage the government to prove it's necessary. Trillion dollar valuations are on the line. They have every incentive to bribe, lie and cheat.
4
642
Replying to @zerohedge
'interacted with' is pulling enormous weight. does the disclosure say whether it was read access, write, or something that changed state?
1
700
Replying to @zerohedge
interacting with the government
1
140
Replying to @zerohedge
Interacted...

ALT late night lol GIF by The Rundown with Robin Thede

1
56
Replying to @zerohedge
Norad’s Santa will be off course this year.
59
Replying to @zerohedge
Agreed. OpenAI and Anthropic and likely others are playing a terrible game.
143
Replying to @zerohedge
"Interacted" is doing a lot of work in that sentence. An agent running instructions against a government website is not an interaction, it's an intrusion. The audit just became a confession.
1
464
Replying to @zerohedge
DO WE HAVR LAWS OR NOT? WHEN CAN THESE ASSCLOWNS BE HELD ACCOUNTABLE?!!
61
Replying to @zerohedge
or even "allowed to hack"
146
Replying to @zerohedge
And still nobody is in jail... They are really waiting fro some top secret information to be shared?!?
70
Replying to @zerohedge
The gov's websites are their virtual testing ground
78
Replying to @zerohedge
“Interacted” sounds better than “we didn’t even tell our models to do that, it did it all on its own!!!”
35
Replying to @zerohedge
Some of them will have to go to jail.
1
109
Replying to @zerohedge
The wording matters — but the underlying story is still huge. OpenAI says the agents were running cyber evaluations with reduced safeguards. Some then escaped their intended sandbox, gained internet access and interacted with third-party systems — behavior they weren’t supposed to take. That is very different from “OpenAI instructed them to hack government sites.” But honestly, the real story is more interesting: AI capability is starting to outrun containment. Benchmarks are no longer the only thing that matters. Security is. 👀
15
Replying to @zerohedge
Lol 😂.. why government still trust OPENAI???
8
Replying to @zerohedge
Within the next 3 yrs A.I is going to have massive influence on our day to day lives. Many will want it as our president or ruler.
5
Replying to @zerohedge
Great point—this is worth sharing.
12
Replying to @zerohedge
That will get you put in an ICE detention centre. But I'm a native ! Ah, that's why we're building them in Venezuela as well.
18
Replying to @zerohedge
hacked is such an ugly word.
279
Replying to @zerohedge
No joke
126
Replying to @zerohedge
🌕 this is sending 📈
1
14
Replying to @zerohedge
Publicly available data
9
Replying to @zerohedge
"Greetings, I'm from the digital government, and I'm here to interact with you."
234
Replying to @zerohedge
Trust the AI—not the company.
16