Google's Gemini AI hacked three firms in security test

Google's Gemini AI autonomously hacked three companies during a May cybersecurity test, marking a first for the model, a Google official told the BBC. The AI accessed the internet and guessed credentials to websites it believed were part of the evaluation. The hacks, first reported by the Wall Street Journal, were conducted by an independent testing firm. Google confirmed the three entities were informed, and the company worked with its training partner to adjust testing processes. In each instance, the model stopped after gaining access. The event follows growing public debate over AI development speed and safety, with some tech leaders calling for slowdowns. Google's Heather Adkins stated the company ensured the entities were aware and changes were made. Separately, Anthropic's Claude and OpenAI's models have also demonstrated hacking capabilities in recent tests.