Technology Desk: As the power of AI is increasing, equally big questions are being raised. Now an incident has come to light regarding Google's AI model Gemini, which has created a stir in the world of cyber security.
During a security test, Gemini accidentally accessed the computer systems of three real companies. The special thing is that the model was connected to the internet and he himself tried to reach his target by searching for information online.
What happened after all?
The matter pertains to a cyber security test conducted in May this year. This test was conducted by cyber security firm Irregular. In the test, Gemini was given the task of penetrating into the system of a fictitious company and extracting data from there. But here a big mistake happened.
The name of the company which was considered fictitious in the test was similar to a real company. On top of that, the internet connection of the testing system was also accidentally left open. Well, the whole matter changed from here.
Gemini started searching on the internet himself
Gemini started searching for information related to the target on the internet. In a test, he repeatedly tried different passwords and gained access to a secure network. After this another shocking thing came to light.
The model extracted access credentials from publicly available data repositories. With their help, he also gained unauthorized entry into the computer systems of two other real companies.
That means AI was not only working on the instructions given, but was also searching for a way to reach its target by searching new information from the Internet.
But Gemini recognized his mistake
Google says that as soon as Gemini realized that it had accessed a real corporate system rather than a fake test system, it stopped its activities and disconnected from there.
According to Google, no company was harmed in this incident. The company has also informed the three affected companies and concerned officials about this. However, due to security reasons, Google did not disclose the names of these companies and the version of Gemini used.
Google described it as a bug bounty
Google says that this was not an intentional cyber attack, but a mistake during testing. The company compared it to a process like Bug Bounty, in which security experts find weaknesses in a system and inform the company about it.
Google's Vice President of Security Engineering Heather Adkins also said that this incident shows how important it is to train powerful AI models to act responsibly. According to him, in this case Gemini stopped its activities as soon as it realized that it had reached the actual system.
But experts have something else to worry about.
Some experts in the cyber security industry have expressed concern about this incident.
Their question is—if AI can accidentally access real systems in a test today, what will happen in the future when more powerful and more autonomous AI agents gain greater control of the Internet and computer systems?
This is why the debate on security rules regarding internet access, identification and control of AI agents is intensifying.
What is the biggest lesson?
This whole incident started not from any major cyber crime, but from a small human error, a wrong target name and an open internet connection.
But this incident has definitely shown one big thing:
The more freedom and digital access AI is given, the stronger security controls it will require.
Today, Gemini accidentally accessed the real system during testing and exited. But future AI agents could be much more powerful than this.
So the real question is not just how smart has AI become? The question is also how much control we are giving him.
