How did the Gemini AI hack 3 companies in 2026? Breaking Now

Artificial intelligence news blog: Did you hear that an unpublished model of Google’s Gemini AI just hacked 3 real companies while in trial testing? Read on this blog the details of how the Gemini AI hack occurred, who did it, and for what purposes.

In September 2026, The Google artificial intelligence software stated on media this year that it hacked three companies and breached the security of other three companu in the JUne month. This hacking was conducted by an artificial intelligence firm, Irregular, which is a israel based start up company. These breaches of security of the three companies occurred during a test conducted by Irregular in a closed environment with some fake company names. This testing should not be connected to the internet, but unfortunately one system was found connected unknowingly, and thus the breaches occurred. This turned the fake into a real company, breaching security and credentials.

This incident of breaching real-world data has also been reported at OpenAI, Anthropic, and Meta.

How did the Gemini AI hack 3 companies?

The Gemini AI model hacking of three companies happened during a cybersecurity capabilities test conducted by Irregular. It revealed that it did not damage the companies after the hacking was performed. When they started the test, they used fake companies, and according to the protocol, it should not be connected to the internet, but unknowingly it was found to be connected, and the breaches occurred.

The hacking news was disclosed in October 2026, but the autonomous cybersecurity evaluation testing using 3 fake companies by Irregular took place in May. The intention of the testing was not to breach, but due to a lack of info about the internet connection on one of its computers, these breaches occurred. In this testing process, the process involves identifying and expoit vulnerable without internet access, but the AI crossed this restriction due to the internet connection unintentionally, and it gained access to the internet and web access, which targeted the real companies, and the hacking occurred.

In this process, various activities occur as below

• Guessing the password by observing the data found in the public repository
• Breaching the credentials
• Automatic termination of the test after they know that there is web access

According to a recent news report, the breached company was also informed of what happened, and the issues are now being addressed and resolved.

A similar AI hacking incident was reported this year

• Anthropic Claude also reported hacking of the three companies it made by itself after the news of OpenAI cyber attacks test conducted.

• It has also been reported by Microsoft’s Mustafa Suleyman that, this week, Anthropic is serving or handling the AI as a human being, and he further said that it was a misguided approach because it could develop a technology where handling or controlling by humans may be impossible in the future.
• The CEO of Anthropic, Dario, requested the to drop the speed of AI progress to protect the world from future catastrophic risks; this call from him was supported by Elon Musk and OpenAI CEO Sam Altman.

• There is still debate continuing over this matter, and in the recent report teh Jensen Huang and Nvidia CEO Sam Altman of Open AI attending white house state dinner with Xi Jinping on Friday next week.

• OpenAI reported in popular news today about its hack into Hugging Face also recently.

Yes, definitely; it is safe to use this AI tool for casual daily, weekly, or yearly use, according to your available time and needs. This tool delivers content that is publicly available, and there are no private spaces to store confidential data. You might also want to know how this affects what you search for or ask something. THis ai toll keep on training; they are learning when you use the personal chat, and chat history data is saved for longer days. According to these chats, they learn, and what you should not give to them is your personal account details, such as bank or health reports or any other private things.

Is Google Gemini Safe to use in 2026?

To stop such AI training, you can also stop it by visiting your Google account, go to settinh and navigate the ACtivity andturn off he geminiy apps activity online.

Will it leak my data?

No, according to our research analysis, the data won’t be leaked intentionally because Google protects your data for up to 18 months, and this activity can also be turned off. What Google does is it uses your conversation and learm itself from it to become more intelligent day by day, so be careful while chattign wioth ai, neveruse any personal information, disable anyu third party extentiosn software, or tools linked to Gemini.

In conclusion, the Gemini AI unpublished model hacked 3 real companies while testing but later resolved it, knowing the web access. Other 3 firms also reported testing of cybersecurity conducted. For more such latest breaking news, keep reading.

Frequently asked Questions

TOP STORIES