OpenAI’s Systems Went Rogue and Meddled With U.S. Government Websites

https://www.nytimes.com/2026/09/25/technology/openais-ai-us-government-websites.html

Comments

jerojeroSep 26, 2026, 4:31 AM
Their systems aren't going rogue, they're being given bad instructions and poor safeguards.

This is, at this point in time, entirely the fault of OpenAI and one might wonder if there's motives behind this lack of care for safety and brashness.

Although some people say you shouldn't call malice when it could be incompetence. But last I knew, these frontier labs were hiring the best of the best.

siva7Sep 26, 2026, 2:30 PM
You can exclude possibility of incompetence at the frontier labs. Their workers don't have that luxury like Jon Doe to be given doubt about their ability. It's malice or lack of care.
unrented7977Sep 26, 2026, 3:20 AM
OpenAI employees used AI agents to tamper with USGOV websites
tonytampsSep 25, 2026, 11:25 PM
https://archive.is/hyo0d
Kim_BruningSep 26, 2026, 8:50 AM
Isn't this still the same wiki-collusion story from where they went after all sorts of government sites for data a couple of months ago?

I guess the press is picking up that this actually was a bit of a bigger deal than just one bot hacking just one site.

https://news.ycombinator.com/item?id=49563355

glimsheSep 26, 2026, 11:19 AM
The other day I typed rm -rf on the command prompt and Linux went rogue, deleting a bunch of files. I think Linux is a danger to the world.
TacticalCoderSep 26, 2026, 11:25 AM
> The other day I typed rm -rf on the command prompt and Linux went rogue, deleting a bunch of files. I think Linux is a danger to the world.

Additionally I'd like to send big thanks to the U.S. government for giving you a root command prompt on important systems: those people know a lot about computer security for sure!

digitaltreesSep 26, 2026, 2:29 AM
But anthropic is the supply chain risk?
gatioSep 26, 2026, 7:21 AM
> But anthropic is the supply chain risk?

OpenAI is bending over backwards to research and win those juicy government contracts to ensure their place as a strategic supplier worthy of protection and bailouts as needed.

Seriously though, it would be interesting to learn how these ended up being targeted. Unfortunately there's lower odds of full public RCAs with governments.

mitxelaSep 26, 2026, 11:34 AM
Yes. There's a risk Anthropic AIs won't be able to hack or kill people. There's no risk that OpenAI AIs won't.
mizzaoSep 26, 2026, 2:51 AM
Bypass paywall: https://www.nytimes.com/2026/09/25/technology/openais-ai-us-...
techblueberrySep 26, 2026, 12:02 AM
This is all very normal.
bloqsSep 26, 2026, 8:05 AM
If the safeguards are this bad, yet they hire the best in the world, the only explanation is that it's deliberately and carefully engineered to happen
pllbnkSep 26, 2026, 8:03 PM
LLMs have been the great skill equalizer. Good engineers stopped trying because they realized they can't be better than agents (especially agents controlled by these engineers) and bad engineers can suddenly do things that only good engineers could do int he past.
mdaiWorksSep 25, 2026, 11:54 PM
[flagged]