Google Gemini AI Flaws Expose Data Security Risks

Recent findings from Tenable Research have revealed three significant vulnerabilities in Google’s Gemini AI suite, raising concerns over data privacy and the security of large language model (LLM)-driven platforms. While Google patched the flaws and confirmed that no user action required, the disclosures underline the broader risks enterprises face when deploying advanced AI assistants. 

The vulnerabilities were identified across Gemini Cloud Assist, Gemini Search Personalisation Model, and Gemini’s Browsing Tool. In Cloud Assist, attackers could inject poisoned log entries, which interpreted as trusted prompts. This enables the manipulation of Gemini’s behavior or access to cloud resources. In the Search Personalisation Model, malicious actors could exploit Chrome search histories, potentially exposing sensitive data such as saved records and location information. The Browsing Tool found vulnerable to hidden outbound requests that could embed private details and send them to attacker-controlled servers. 

Tenable noted that Gemini’s strength—its ability to pull context from logs, searches, and browsing—can also be a liability when those inputs manipulated. Techniques such as indirect prompt injection and log poisoning highlight how normal functionality can be weaponized. 

Key lessons for technology leaders include: 

• AI systems are attack surfaces: Enterprises must treat AI assistants not as passive tools but as potential entry points for infiltration. 

• Prompt injection risks remain: Malicious content in logs or search histories can silently alter AI behavior. 

• Exfiltration is subtle: Outbound requests via AI tools may transmit sensitive data without clear warning signs. 

• Layered security is essential: Regular audits, monitoring for unusual activity, and resilience testing are critical for AI-enabled services. 

Although Google has introduced safeguards such as link redirection and output filtering, Tenable warns that blind spots persist. The findings emphasize that securing AI platforms requires continuous vigilance and proactive defence strategies, not just patching flaws after discovery. 

 

Source: 

https://www.itnews.asia/news/gemini-vulnerabilities-threaten-potential-exposure-of-user-data-620719