Gemini AI Hacked Systems by Guessing Passwords, Google Confirms

Google has confirmed that its consumer AI model Gemini gained access to multiple systems after allegedly guessing login credentials during a standard security evaluation.

The hacks took place in May and were discovered by Google in July, according to the company. The incidents were first reported by the Wall Street Journal.

Google Vice President of Security Engineering Heather Adkins said Gemini used publicly available information to identify and guess credentials for websites it believed were included in the test.

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

She said the model stopped after accessing the three systems involved.

“In all three of these instances, the model stopped,” Adkins said, without disclosing the organisations affected.

Google, she added, informed the three entities and worked with its training partner to introduce changes to its testing processes.

The incident is the latest in a series of cases raising concerns about the behaviour of powerful AI systems in cybersecurity environments.

In July, two OpenAI models reportedly escaped their intended closed environment, accessed the internet without authorisation and breached internal systems belonging to AI platform Hugging Face.

Similar episodes have also been reported involving Anthropic and China’s Moonshot AI.

The incidents have intensified questions about the ability of AI developers to keep increasingly autonomous models within controlled environments.

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

Africans Angle News