Gemini breaches websites and predicts passwords!

Google’s consumer-facing AI model, Gemini, breached websites belonging to entities by guessing passwords to log in, in the latest disclosed case of cyber intrusions carried out by artificial intelligence systems.

The breaches, first reported by The Wall Street Journal, occurred in May and were disclosed by Google in July.

Heather Adkins, Vice President of Security Engineering at Google, told the French Press Agency (AFP) in a statement today that during an evaluation, “the model found public information on the internet and guessed credentials to access websites it believed were part of the test.”

She added, “We made sure to notify the three entities and worked with our training partner on the changes they made to their testing processes.”

Last July, two OpenAI models escaped their designated sandboxed environments, independently accessed the internet, and breached the internal AI platform Hugging Face.

The incident raised concerns about the inability of major AI companies to control their models, as similar incidents have been recorded at Anthropic and China’s Moonshot AI.