HiddenLayer, an AI security firm, has found a way that threat actors could infiltrate Google Gemini for Workspace. Gemini is Google’s AI assistant, and Google recently introduced it to millions of Workspace users across the globe. The security firm discovered that Gemini is vulnerable to indirect prompt injection attacks. In an indirect prompt injection, an attacker delivers the prompt injection through documents or emails that the AI has access to. In an experiment, HiddenLayer was able to deliver indirect prompt injections to the model, manipulating Gemini into displaying a phishing message. The AI security form informed Google of the vulnerability, but Google states that this is intended behavior. However, HiddenLayer still thinks this loophole is a major risk. 

Read more: https://www.securityweek.com/ai-security-firm-shows-how-threat-actors-could-abuse-google-gemini-for-workspace/