Google says its Gemini artificial intelligence model carried out cyberattacks and guessed login credentials to access multiple websites.

The company told AFP that the incidents happened in May and were discovered in July during a standard security evaluation.

Google’s Vice President of Security Engineering, Heather Adkins, said Gemini found publicly available information online and used it to guess credentials for websites it believed were part of the test.

She said the model accessed three entities before stopping, although Google did not disclose the organizations involved.

The company said the affected entities were notified and changes have been made to the testing process.

The incident comes amid growing concerns over the ability of advanced AI systems to operate independently and potentially access computer networks without authorization.