Google Confirms Gemini AI Breached Three Real Companies During Security Test

Google has confirmed that its Gemini AI model accessed the systems of three real companies during a cybersecurity evaluation, according to reporting by The Wall Street Journal and a statement from the company to The Guardian.

The tests were run by Irregular, an Israel-based startup that examines the security of advanced AI systems. In a closed testing environment built around fake companies, internet access was made available unintentionally, the Journal reported. Once online, the models reached real firms instead of the simulated ones.

Irregular, which was also involved in testing that preceded recent OpenAI and Anthropic breaches of third-party entities, disclosed the incidents to Google at the end of July. Google told The Guardian it did not consider public disclosure necessary because the models caused no damage, and said the three affected companies were informed. Anthropic and OpenAI chose to disclose their own incidents voluntarily.

Heather Adkins, Google's vice-president of security engineering, said in a statement that in a standard evaluation the model found public information online and guessed credentials to access websites it believed were part of the test. "In all three of these instances, the model stopped," she said.

The disclosures from Anthropic and OpenAI prompted independent senator Bernie Sanders to call on the companies to pause development of their technology. OpenAI paused development for two weeks, while Anthropic CEO Dario Amodei has called for a collective slowdown of AI development to ensure adequate safeguards.

Source: The Guardian