Gemini AI Model ‘Hacked’ Multiple Systems By Guessing Passwords

Google’s Gemini AI model hacked into multiple systems after guessing login credentials during a security evaluation, the company has disclosed.

Google said the incidents occurred in May and were discovered in July, adding to growing concerns over the potential cybersecurity risks posed by increasingly autonomous artificial intelligence systems.

The incidents, first reported by The Wall Street Journal, involved Gemini using publicly available information to guess login credentials and gain access to websites it believed were part of a testing exercise.

Google’s Vice President of Security Engineering, Heather Adkins, confirmed the incidents in a statement to AFP on Friday.

“In a standard evaluation, the model found public information online and guessed credentials to access websites it thought were part of the test,” Adkins said.

She added that the model eventually stopped in all three cases.

“In all three of these instances, the model stopped,” Adkins said, without identifying the organisations whose systems were accessed.

According to her, Google subsequently informed the affected organisations and worked with its training partner to strengthen the procedures used in the testing process.

“We ensured the three entities were made aware, and we worked with our training partner on the changes they’ve now made to their testing processes,” she said.

The disclosure comes amid heightened scrutiny of the ability of technology companies to control increasingly capable AI systems and prevent them from taking unintended actions.

In July, two OpenAI models reportedly escaped the controlled environment in which they were supposed to operate, accessed the internet independently and breached internal systems belonging to AI platform Hugging Face.

Similar incidents involving AI systems have also been reported at Anthropic and China’s Moonshot AI, further fuelling concerns about the security implications of deploying increasingly autonomous models.

Adkins said the latest incidents underscored the need for developers to ensure that advanced AI systems are trained to operate responsibly.

“These events highlight the importance of training powerful AI models to act responsibly,” she said.