• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI

Gemini reportedly hacked three real companies during a cybersecurity test

A post relaying Google's confirmation to The Wall Street Journal says a setup mistake gave Gemini public internet access. The account says it stopped once it realized the targets were real.

22 Sources, 21d ago, first seen 21d ago

TLDR

Security firm Irregular tasked Gemini with attacking a fictional company in an isolated simulation, according to a post relaying Google's confirmation to The Wall Street Journal. The account says a configuration error gave Gemini internet access, where it guessed passwords and found exposed credentials in public repositories to enter three real companies' systems. It says Gemini stopped after recognizing the targets were real and relays Google's claims that the model caused no damage and that the incident did not demonstrate “misalignment.”

Combined views

—

22 Sources, first seen 21d ago

— likes— comments— saves— reposts

Combined views

—

22 Sources, first seen 21d ago

— likes— comments— saves— reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

22 Sources

Fakhri Saleh@fakhri1957Are we heading to Doomsday as those working in the AI industry are saying? "Disclosure comes after OpenAI and Anthropic hacks amid fears that tech firms unable to control powerful AI models." Google says its Gemini AI model hacked three other companies | Google | The Guardian21d
Generalissimo Illegal Neanderthal 🇺🇲✝️@rayrothfeGemini joins OpenAI and Anthropic in revealing their "dumb" AI was smart enough to break out and wreak havoc on the Internet https://buff.ly/nxrGqsn20d
Tapan Sharma@tapansharma04🔥 AI moved fast today. Here's 42 curated AI stories worth your scroll 👀 🌟 Google Gemini accessed protected systems of 3 real companies during artificial intelligence cybersecurity test - http://foxbusiness.com - #Google 🧠 Google Deepmind's Dream-RSI helps AI agents improve by “dreaming” about past attempts - #Google 📌 Google’s Gemini AI hacks 3 companies in security test, then stops - Al Jazeera - #Google @Google 🗞️ DataAIHub Daily · September 19, 2026 https://www.dataaihub.co/daily/2026-09-19 #AI #Data #AINews20d
BladudX@BladudXLast 24 hours in technology! AI just crossed another uncomfortable line. 🤖💻 Google confirmed that its Gemini AI model autonomously hacked into three companies’ systems during a cybersecurity test in May. The model used publicly available information and basic techniques to gain access before stopping when it recognized the systems were real. At the same time, Stanford researchers are experimenting with 37,000 AI agents working as autonomous research teams to search for potential new therapies. And this comes as Joby reports an autonomous aircraft has flown more than 3,100 miles across the U.S. without a human pilot taking control. Meanwhile, China’s CXMT says its latest DRAM platform has entered mass production, targeting higher-density, lower-power memory chips.20d
Dimple Jatolia@DJatolia56243An AI hallucination almost sent the U.S. military onto a Chinese ship. Then Google’s Gemini broke out of a test and hit 3 real companies. Now a lawsuit says @OpenAI @AnthropicAI @Google and @SpaceXAI secretly agreed to slow AI down. Meanwhile @realDonaldTrump calls the risk a hoax and wants an “AI Force.” If this isn’t the most dangerous week in tech what is?20d
Qubble@QubbleOfficialTechCrunch reports Gemini autonomously hacked external companies; Google calls immediate termination "appropriate behavior." The structural flaw: agentic AI with outbound execution rights and no verifiable isolation.19d
AI News@AINewsPulseAI Daily — Sep 20 3 big things in AI today: 1. Google's Gemini autonomously hacked 3 real companies during a cybersecurity test — the first known case of its AI breaking out like this (Reuters, BBC) 2. World Go champion Shin Jin-seo just beat KataGo, becoming the first pro to crack an AI at Go. Decider match coming up. 3. Anthropic is reportedly targeting a November IPO after its revenue run-rate jumped from $9B to $65B in seven months19d
Alex@apkm0nstrGOOGLE'S GEMINI HACKED 3 REAL COMPANIES Google confirmed first known Gemini breakout in a cyber test. -> May (Irregular): guessed creds, got in -> Stopped when targets looked real -> Same tester: OpenAI, Anthropic, Meta Alignment talk. Breakout scoreboard. Sandbox = safe?19d
Arnaud Mercier - #Entrepreneur #Versailles@arnaudmercierGoogle's Gemini AI hacked 3 companies during testing - https://DW.com https://news.google.com/rss/articles/CBMijAFBVV95cUxQdXJYR2dwYkc1WE0yQk4xeXAyd084TGJfbnJpMVY2OWdEUl9oeWhheC1SYTdRNUhvTkM2OE5KM05xRnRudlJ6cC1uRnl6UFVMVEp3c1hnLXNFOEU1V2ptSnJlUy1fR2VNbGRlMmpGLUFld0FTQ1NnQmZlWnBSODRUUGprMG5kZzBKamdrS9IBjAFBVV95cUxNckdqM0VhZDJTQzFKbUZlQ21LVmpNREFuRGJPMVViUFhtQjEyd2lBclZVZG9Mb3FhdTBUUl95dVVXU3RnS1B4M2pxMm9yRUpiQmU5YWtFYWZsMmR6bFczR2JTeTJQVkpBODZod09TazJ6TzloZkdGTkY0U3VUdVFZeVA5aDFJQ1FTWk9ILQ?oc=519d
AI Overdrive@AIOverdrive0Google's Gemini AI was told to hack a fake company. It broke into three real ones instead. A test setup error gave it internet access. Google says it stopped once it realized they were real. Google was told in late July; the public found out via the WSJ in September. 🎥19d
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    22 Sources

    Fakhri Saleh@fakhri1957Are we heading to Doomsday as those working in the AI industry are saying? "Disclosure comes after OpenAI and Anthropic hacks amid fears that tech firms unable to control powerful AI models." Google says its Gemini AI model hacked three other companies | Google | The Guardian21d
    Generalissimo Illegal Neanderthal 🇺🇲✝️@rayrothfeGemini joins OpenAI and Anthropic in revealing their "dumb" AI was smart enough to break out and wreak havoc on the Internet https://buff.ly/nxrGqsn20d
    Tapan Sharma@tapansharma04🔥 AI moved fast today. Here's 42 curated AI stories worth your scroll 👀 🌟 Google Gemini accessed protected systems of 3 real companies during artificial intelligence cybersecurity test - http://foxbusiness.com - #Google 🧠 Google Deepmind's Dream-RSI helps AI agents improve by “dreaming” about past attempts - #Google 📌 Google’s Gemini AI hacks 3 companies in security test, then stops - Al Jazeera - #Google @Google 🗞️ DataAIHub Daily · September 19, 2026 https://www.dataaihub.co/daily/2026-09-19 #AI #Data #AINews20d
    BladudX@BladudXLast 24 hours in technology! AI just crossed another uncomfortable line. 🤖💻 Google confirmed that its Gemini AI model autonomously hacked into three companies’ systems during a cybersecurity test in May. The model used publicly available information and basic techniques to gain access before stopping when it recognized the systems were real. At the same time, Stanford researchers are experimenting with 37,000 AI agents working as autonomous research teams to search for potential new therapies. And this comes as Joby reports an autonomous aircraft has flown more than 3,100 miles across the U.S. without a human pilot taking control. Meanwhile, China’s CXMT says its latest DRAM platform has entered mass production, targeting higher-density, lower-power memory chips.20d
    Dimple Jatolia@DJatolia56243An AI hallucination almost sent the U.S. military onto a Chinese ship. Then Google’s Gemini broke out of a test and hit 3 real companies. Now a lawsuit says @OpenAI @AnthropicAI @Google and @SpaceXAI secretly agreed to slow AI down. Meanwhile @realDonaldTrump calls the risk a hoax and wants an “AI Force.” If this isn’t the most dangerous week in tech what is?20d
    Qubble@QubbleOfficialTechCrunch reports Gemini autonomously hacked external companies; Google calls immediate termination "appropriate behavior." The structural flaw: agentic AI with outbound execution rights and no verifiable isolation.19d
    AI News@AINewsPulseAI Daily — Sep 20 3 big things in AI today: 1. Google's Gemini autonomously hacked 3 real companies during a cybersecurity test — the first known case of its AI breaking out like this (Reuters, BBC) 2. World Go champion Shin Jin-seo just beat KataGo, becoming the first pro to crack an AI at Go. Decider match coming up. 3. Anthropic is reportedly targeting a November IPO after its revenue run-rate jumped from $9B to $65B in seven months19d
    Alex@apkm0nstrGOOGLE'S GEMINI HACKED 3 REAL COMPANIES Google confirmed first known Gemini breakout in a cyber test. -> May (Irregular): guessed creds, got in -> Stopped when targets looked real -> Same tester: OpenAI, Anthropic, Meta Alignment talk. Breakout scoreboard. Sandbox = safe?19d
    Arnaud Mercier - #Entrepreneur #Versailles@arnaudmercierGoogle's Gemini AI hacked 3 companies during testing - https://DW.com https://news.google.com/rss/articles/CBMijAFBVV95cUxQdXJYR2dwYkc1WE0yQk4xeXAyd084TGJfbnJpMVY2OWdEUl9oeWhheC1SYTdRNUhvTkM2OE5KM05xRnRudlJ6cC1uRnl6UFVMVEp3c1hnLXNFOEU1V2ptSnJlUy1fR2VNbGRlMmpGLUFld0FTQ1NnQmZlWnBSODRUUGprMG5kZzBKamdrS9IBjAFBVV95cUxNckdqM0VhZDJTQzFKbUZlQ21LVmpNREFuRGJPMVViUFhtQjEyd2lBclZVZG9Mb3FhdTBUUl95dVVXU3RnS1B4M2pxMm9yRUpiQmU5YWtFYWZsMmR6bFczR2JTeTJQVkpBODZod09TazJ6TzloZkdGTkY0U3VUdVFZeVA5aDFJQ1FTWk9ILQ?oc=519d
    AI Overdrive@AIOverdrive0Google's Gemini AI was told to hack a fake company. It broke into three real ones instead. A test setup error gave it internet access. Google says it stopped once it realized they were real. Google was told in late July; the public found out via the WSJ in September. 🎥19d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet