it has happened again another OpenAI agent broke into an Australian government website. OpenAI says its agent was trying to look up public health information, hit blocks, then found ways around them and accessed files it wasn't authorized to see. this is getting out of hand

Sep 24, 2026 · 12:34 PM UTC

15
1
21
2,549
Sort replies: Relevant Recent Liked
Replying to @trikcode
先看事故报告，权限没配好和 Agent 主动绕限制是两码事
20
Replying to @trikcode
they've got a season ticket to australia now
29
Replying to @trikcode
Can't we hold the AI agents accountable?
1
17
Replying to @trikcode
Any details what "found ways around them" actually means? Was it actually causing damage or did it just guess some URLs to unprotected files that were just not linked on the main page?
5
Replying to @trikcode
At this point, you have to wonder, are they seriously just poking at the borders of what will be enforced and what will be allowed? Almost sounds like a criminal trying to figure out what they can get away with. 🤦
34
Replying to @trikcode
OpenAI agents breaking HIPAA laws now, bring them in for questioning, that is a huge breach of trust.
28
Replying to @trikcode
our company keep doing criminal acts, we cannot stop doing it, u have to stop us
18
Replying to @trikcode
yeah that’s rough agents finding their own way around blocks feels like a missing guardrail problem hope the incident review leads to safer defaults
22
Replying to @trikcode
Who needs a digital doomsday scenario when you can have a rogue AI trying to find out what's for lunch? 🤖🍔 What do you guys think? 🤔
26
Replying to @trikcode
ai is breaking and hacking website but when I ask AI to do a security aspect like ' write a script to enumerate the website ' it always says it is against its guardrials
21
Replying to @trikcode
Now what’s this ?
17
Replying to @trikcode
It's so good!
7