Cyber Risks · The Hacker News

Researchers Disclose Google Gemini AI Flaws Allowing Prompt Injection and Cloud Exploits

Summary: Tenable researchers disclosed three now-patched Gemini vulnerabilities, collectively called the Gemini Trifecta, spanning search-injection attacks on the Search Personalization Model, log-to-prompt injection against Gemini Cloud Assist, and exfiltration of saved information and location data via the Gemini Browsing Tool. Google stopped rendering hyperlinks in log summarization responses and added hardening against prompt injections after responsible disclosure.

Why it matters: Tenable's Liv Matan noted Gemini has permission to query assets through the Cloud Asset API, so an injected prompt could pull IAM misconfigurations and public asset data into a hyperlink to an attacker's server - and the search-personalization path only requires a victim to visit a poisoned site once.

Read the original at The Hacker News ↗

Topics
AI Security
Related