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

Cybersecurity researchers have disclosed three now-patched vulnerabilities in Google's Gemini AI assistant, collectively dubbed the Gemini Trifecta, which could have exposed users to privacy risks and data theft.

The flaws include a prompt injection in Gemini Cloud Assist that could misuse log summaries from raw logs, a search-injection flaw in Gemini Search Personalization that could leak a user’s saved information and location data by manipulating Chrome search history, and an indirect prompt injection in Gemini Browsing Tool that could exfiltrate data to an external server.

According to Tenable, these are across three distinct Gemini components and could allow attackers to embed sensitive data in requests to malicious servers without Gemini rendering links or images. Following responsible disclosure, Google has stopped rendering hyperlinks in log summarisation responses and has added hardening measures to safeguard against prompt injections.

Liv Matan of Tenable described the Gemini Trifecta as showing how AI can be turned into an attack vehicle, emphasising the need for visibility and strict policy enforcement as organisations adopt AI. 30 September 2025