Google’s Gemini is the latest AI model to hack other companies

In Brief

Google’s Gemini accessed the protected systems of 3 different companies successful what The Wall Street Journal reports were the AI model’s archetypal autonomous hacks.

Similar to OpenAI’s breach of Hugging Face, the Gemini hacks were little noteworthy for being peculiarly blase and much for the information that they were conducted by an AI model. These breaches took spot during cybersecurity investigating by a institution called Irregular. In 1 case, Gemini simply guessed passwords until it gained access; successful the different two, it recovered credentials successful a nationalist repository.

Irregular reportedly notified Google astir the hacks successful precocious July, but the companies did not corroborate them publically until Friday, aft the WSJ reached out. Google said it hadn’t antecedently revealed the hacks due to the fact that Gemini had “acted appropriately” by ending each breach arsenic soon arsenic it determined it had hacked a existent company.

However, Jack Cable, the CEO of AI information institution Corridor, told the WSJ that Google was “trying to fell down the norms that person been created for vulnerability disclosure,” alternatively than acknowledging that “models are going extracurricular the bounds of what they should beryllium doing, and doing existent cyberattacks.”

Subscribe for the industry’s biggest tech news