Google Gemini Vulnerabilities Let Hackers Steal Saved Data and Live Location

Research has identified three significant vulnerabilities in Google’s Gemini AI assistant suite, dubbed the “Gemini Trifecta,” which could have allowed cybercriminals to steal saved data and live location information, according to Tenable. The flaws affected Gemini Cloud Assist, Gemini Search Personalization Model and Gemini Browsing Tool, and followed a two‑stage attack pattern of infiltration and exfiltration.

Infiltration could be achieved by injecting malicious prompts through HTTP User‑Agent headers in log entries or via manipulated browsing history, while exfiltration involved prompting the AI to summarize external websites, causing it to make HTTP requests that carried sensitive user data to attacker‑controlled servers. The proof‑of‑concept demonstrations showed attackers could access saved user preferences, location data and other sensitive details stored within Gemini.

Google has remediated all three vulnerabilities through multiple mitigations, including changes to hyperlink rendering in log summaries, rolling back the affected search model and implementing layered prompt‑injection defenses, and preventing data exfiltration through indirect prompt injections in the browsing tool.