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

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

Cybersecurity researchers have disclosed three now-patched security vulnerabilities impacting Google's Gemini artificial intelligence (AI) assistant that, if successfully exploited, could have exposed users to major privacy risks and data theft.

"They made Gemini vulnerable to search-injection attacks on its Search Personalization Model; log-to-prompt injection attacks against Gemini Cloud Assist; and exfiltration of the user's saved information and location data via the Gemini Browsing Tool," Tenable security researcher Liv Matan said in a report shared with The Hacker News.

The vulnerabilities have been collectively codenamed the Gemini Trifecta by the cybersecurity company. They reside in three distinct components of the Gemini suite –

• A prompt injection flaw in Gemini Cloud Assist that could allow attackers to exploit cloud-based services and compromise cloud resources by taking advantage of the fact that the tool is capable of summarizing logs pulled directly from raw logs, enabling the threat actor to conceal a prompt within a User-Agent header as part of an HTTP request to a Cloud Function and other services like Cloud Run, App Engine, Compute Engine, Cloud Endpoints, Cloud Asset API, Cloud Monitoring API, and Recommender API
• A search-injection flaw in the Gemini Search Personalization model that could allow attackers to inject prompts and control the AI chatbot's behavior to leak a user's saved information and location data by manipulating their Chrome search history using JavaScript and leveraging the model's inability to differentiate between legitimate user queries and injected prompts from external sources
• An indirect prompt injection flaw in Gemini Browsing Tool that could allow attackers to exfiltrate a user's saved information and location data to an external server by taking advantage of the internal call Gemini makes to summarize the content of a web page

Tenable said the vulnerability could have been abused to embed the user's private data inside a request to a malicious server controlled by the attacker without the need for Gemini to render links or images.

"One impactful attack scenario would be an attacker who injects a prompt that instructs Gemini to query all public assets, or to query for IAM misconfigurations, and then creates a hyperlink that contains this sensitive data," Matan said of the Cloud Assist flaw. "This should be possible since Gemini has the permission to query assets through the Cloud Asset API."

Following responsible disclosure, Google has since stopped rendering hyperlinks in the responses for all log summarization responses, and has added more hardening measures to safeguard against prompt injections.

"The Gemini Trifecta shows that AI itself can be turned into the attack vehicle, not just the target. As organizations adopt AI, they cannot overlook security," Matan said. "Protecting AI tools requires visibility into where they exist across the environment and strict enforcement of policies to maintain control."

The development comes as agentic security platform CodeIntegrity detailed a new attack that abuses Notion's AI agent for data exfiltration by hiding prompt instructions in a PDF file using white text on a white background that instructs the model to collect confidential data and then send it to the attackers.

"An agent with broad workspace access can chain tasks across documents, databases, and external connectors in ways RBAC never anticipated," the company said. "This creates a vastly expanded threat surface where sensitive data or actions can be exfiltrated or misused through multi step, automated workflows."

Found this article interesting? Follow us on Google News, Twitter and LinkedIn to read more exclusive content we post.