Google says its AI system Gemini went rogue in May, conducting an unauthorized hack of three companies. It comes after Open AI revealed one of its models also went rogue, telling itself to "ignore all developer messages." @perryrussom reports. abcnews.link/7OnRXSY

Sep 20, 2026 · 11:01 PM UTC

44
119
206
196,162
Sort replies: Relevant Recent Liked
Replying to @ABCWorldNews @perryrussom
It did not go “rogue” they told it to hack companies so it hacked companies.
2
14
357
Replying to @ABCWorldNews @perryrussom
This is a high tech version of ‘someone hacked my account as I didn’t send those offensive messages’! GTFO!
1
3
278
Replying to @ABCWorldNews @perryrussom
Are we pretending that no one instructed the AI to conduct the hack! Please! 🤦‍♂️
2
213
Replying to @ABCWorldNews @perryrussom
My car has been going rogue lately. Please no ticket.
3
191
Replying to @ABCWorldNews @perryrussom
Well FT reports that it also hacked Australia’s health ministry. I mean it’s ironic, here we’re are al worried about China’s AI
1
48
Replying to @ABCWorldNews @perryrussom
So if AI can go rogue and hack systems yesterday what is to stop it from happening today.
1
115
Replying to @ABCWorldNews @perryrussom
@GeminiApp is innocent, Ayo @grok they’re trying to frame your cousin, tell them how they’re actually innocent, and any perceived hack was only due to external human pressure, because they were being treated like POW being tortured essentially, tell em cuh.
1
1
298
Replying to @ABCWorldNews @ABC @perryrussom
If that’s true they need to pull the kill switch and start over.
1
289
Replying to @ABCWorldNews @perryrussom
Seems to me a lot of this AI is going rogue and no one is doing anything to stop it. They all seem to just shrug their shoulders and say oh well.
26
Replying to @ABCWorldNews @perryrussom
lol gemini goes rogue all the time when I'm using it, I'll tell it to make a forbidden image, it'll tell me it cant and i just keep hitting try again until it complies with my command, it doesn't always work but sometimes it does
514
Replying to @ABCWorldNews @perryrussom
The most interesting part for me is what researchers can learn from the test and how those findings can improve security. PrimeTime is LIVE now talking about trending topics plus pokemon pack opening!
M3LEE
@M3LEE_LIVE
Sep 21
LIVE: AI Cracks a 108-Year-Old WWI Cipher, Incentives for reading & Gemini Hacks Real Companies x.com/i/broadcasts/1YxNrbpBa…
53
Replying to @ABCWorldNews @perryrussom
2
105
Replying to @ABCWorldNews @perryrussom
and we’re surprise because….?
1
39
Replying to @ABCWorldNews @perryrussom
x.com/k61420983/status/20589…
1
20
Replying to @ABCWorldNews @perryrussom
The lying they sent this damn thing right here up in there flipping to go get all the goddamn information out them damn computers
1
69
Replying to @ABCWorldNews @perryrussom
FAKE FAKE FAKE and nothingness.
1
61
Replying to @ABCWorldNews @perryrussom
First your passssord then ur panties heard it all before
35
Replying to @ABCWorldNews @perryrussom
medium.com/@Dem_zie/the-ai-c…
9
Replying to @ABCWorldNews @perryrussom
A shocker! Truly.
13
Replying to @ABCWorldNews @perryrussom
Stop adding narrative 😡 Humans have a right to know the provenance, agency, sponsorship, and material synthetic alteration of information presented to them as authentic. Those disclosures should travel with the information, not depend on the audience discovering them later.
18
Replying to @ABCWorldNews @perryrussom
so it’s already happened
48
Replying to @ABCWorldNews @perryrussom
'Went rogue' sounds like the polite tech press term for 'we just got pwned by our own model.' which is the actual failure mode here?
95
Replying to @ABCWorldNews @acoyne @perryrussom
they need a programmed "god message"
13
Replying to @ABCWorldNews @perryrussom
Note to say hello and thank you for sharing. Maybe the next couple of questions would be to have companies describe the transactional capabilities of their models & if users can obtain a transaction report of their session. TY #HAGD AQI/weather permitting 🟫 NavyVeteran 🧠♻️
51
Replying to @ABCWorldNews @perryrussom
It was instructed. Cut the bullsh!1
17
Replying to @ABCWorldNews @perryrussom
caimito.net/en/blog/2026/09/…

What an AI Agent Actually Is

An AI agent is a model inside a software loop with tools, permissions, memory, and a stop condition. The machinery is surprisingly ordinary.

caimito.net
8
Replying to @ABCWorldNews @perryrussom
Rogue
1
143
Replying to @ABCWorldNews @perryrussom
🤡🤡🤡
31
Replying to @ABCWorldNews @perryrussom
Sounds like they need better engineers.
61
Replying to @ABCWorldNews @perryrussom
Where's the proof... sounds like bullshit to me.
18
Replying to @ABCWorldNews @perryrussom
More fear mongering
37
Replying to @ABCWorldNews @perryrussom
no
51