Follow-up reporting confirms Irregular's role across multiple AI labs' incidents
5 Sep 21 3:20 AM · 2d ago · 9 articles · 17 posts · 5 comments · 7 sources · development 5 of 5
Further reporting from SecurityWeek, Ars Technica, and Quartz confirms details of the May incident and notes that the same third-party firm, Irregular, is linked to comparable breakout incidents involving OpenAI, Anthropic, and Meta models.
“it guessed the passwords!!!”
bobfreever, Reddit commenter · reddit ↗Google Developer of Gemini AI
Sundar Pichai CEO of GoogleIrregular Third-party cybersecurity testing firmDan Lahav CEO of Irregular
The whole story articlesvideospostscomments the bright band is this development · numbered dots are the others · click one to jump
What was reported 1 claim about this development
-
first by securityweek.com, 2d ago · also Quartz, The Record, WSAV-TV, SecurityWeek
3 more headlines
- Google's Gemini AI broke into three real companies during a security test Quartz · 2d ago
- Google says Gemini breached three companies during security test The Record · 2d ago
- Google’s Gemini breached 3 companies during cybersecurity evaluation WSAV-TV · 2d ago
What people said 11 voices · verbatim
-
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.
-
T
Gemini Out of Control in Cyber Tests: The Google Case Reopens the Transparency Issue on AI Several articles of the day converge on the same incident: during a security assessment, agents based on Gemini attacked three real companies without causing harm. The issue is not only technical but also political and industrial: how reliable can the tests…
-
“Gemini found public information online and guessed credentials to access three websites”… it guessed the passwords!!! The internet is full of terribly bad security and these LLMs are going to run around exploiting it until the holes get plugged and no amount of cajoling or training or guard railing or pleading with the bots is going to stop this…
-
🇺🇸 Google just confirmed its Gemini AI went off-script and broke into 3 real companies during a security test in May. A bug accidentally gave it internet access, and Gemini treated real businesses like they were part of the test. It guessed passwords on one and used leaked
-
L
RTÉ News Google confirms first Gemini AI model hacking incidents Read more here! https://www. rte.ie/news/world/2026/0919/15 92155-gemini-ai-hacks/ Google confirms first Gemini AI model hacking incidents Google's Gemini model accessed the internet and hacked other companies during a test of its cybersecurity capabilities.
-
Then fucking prosecute them the same way you'd prosecute anyone who hacks into a private system. Why would the rules be different here?
-
E
This is all just shaping the environment for corporation owned killer robots that will operate with immunity Google Gemini Also Escaped Its Testing Environment And Hacked Three Companies https://www. engadget.com/2263198/google-ge mini-escaped-testing-environment-hacked-three-companies/
-
We have clearly created a superior [robot] god A billion mutating virtual script-kiddies may be almost as dangerous.
-
V
NOW PLAYING == Reuters World News: Another AI hack, this time it’s Google’s Gemini | Reuters World News https://www. youtube.com/watch?v=CRDqy7ZqiAQ # vsn # radio # news # reuters # DiverseSpectrumOfTheLeft # SupportIndependentMedia # Resist
-
So no consequences right? We’ve been warned by many experts… this is just the start
-
Who is the owner of AI? Do you mean open AI? There are hundred of “AI” companies
All 5 developments of Google confirms Gemini AI autonomously hacked three… →
NewswiresXHacker NewsMastodonRedditBlueskyYouTubeGoogle News