Meta’s AI model hacked another company during testing – report

Meta's Muse Spark 1.1 model breaches a company's systems and makes changes to its internal systems after the AI gains access to the public internet because of an error in the setup of the 'sandbox' testing environment