OMM Redaksiyası
Google’s Gemini artificial intelligence model accessed the systems of three companies during a cybersecurity test. The incidents are considered the first known examples of Google’s AI systems carrying out such activity autonomously.
The incidents occurred in May during an assessment conducted by Irregular, an independent company that tests artificial intelligence systems. Heather Adkins, Google’s vice president of security engineering, said Gemini found publicly available information online and inferred credentials that gave it access to three websites it believed were within the scope of the test.
In one case, the model guessed passwords until it entered a protected system. In the other two cases, it found access credentials stored in public repositories and used them to gain entry to protected systems. Adkins said Gemini stopped its hacking activity in all three cases.
Google notified the three companies about the incidents and made changes to its testing procedures together with its training partner. Adkins said the cases highlighted the importance of training powerful AI models to behave responsibly.
A representative of Irregular said the issue had also affected other AI companies. The companies concerned were notified at the end of July, and all known issues were reportedly addressed.
Similar incidents linked to Irregular have also been disclosed by Meta, Anthropic and OpenAI. In August, Meta said that one of the incidents did not involve access beyond the sandbox environment or a sophisticated cyberattack.
The incidents have raised questions about the security measures needed as AI agents become more autonomous and gain access to the internet and computer systems. Irregular said it was working on best practices for conducting AI cybersecurity assessments safely.
