Google Gemini prompt injection and exfiltration flaws (multiple vulnerabilities)
Vulnerability
Summary
Hide ▲
Show ▼
Google Gemini vulnerability disclosure: Tenable identified the Gemini Trifecta across Gemini Cloud Assist, the Search Personalization Model, and the Gemini Browsing Tool. The flaws enabled prompt injection and search-injection paths that could expose saved information, location data, and cloud resources. Google has already patched the issues, including changes to log-summary hyperlink rendering and other hardening against prompt injection.
Related Happenings
Gemini CLI workspace trust RCE flaw
Vulnerability
H score39
First: 30.04.2026 10:07
Last: 30.04.2026 10:07
Sources 1
About this happening:
Google has fixed Gemini CLI and google-github-actions/run-gemini-cli flaws that let untrusted workspace content trigger arbitrary commands on the host in CI/head...
Gemini CLI workspace trust RCE flaw
VulnerabilityAbout this happening: Google has fixed Gemini CLI and google-github-actions/run-gemini-cli flaws that let untrusted workspace content trigger arbitrary commands on the host in CI/head...
Google Antigravity critical prompt-injection RCE flaw
Vulnerability
H score28
First: 21.04.2026 13:52
Last: 21.04.2026 13:52
Sources 1
About this happening:
Google fixed a critical Antigravity flaw that let a prompt injection bypass Secure Mode and escalate to sandbox escape and remote code execution (RCE). The...
Google Antigravity critical prompt-injection RCE flaw
VulnerabilityAbout this happening: Google fixed a critical Antigravity flaw that let a prompt injection bypass Secure Mode and escalate to sandbox escape and remote code execution (RCE). The...
Coruna watering-hole and fake-site exploitation campaign
Campaign
H score44
First: 26.03.2026 13:07
Last: 26.03.2026 13:07
Sources 1
About this happening:
A suspected Russia-aligned nation-state actor is using Coruna in watering-hole attacks in Ukraine and a mass exploitation campaign, expanding the kit’s abuse beyon...
Coruna watering-hole and fake-site exploitation campaign
CampaignAbout this happening: A suspected Russia-aligned nation-state actor is using Coruna in watering-hole attacks in Ukraine and a mass exploitation campaign, expanding the kit’s abuse beyon...
Perplexity Comet prompt-injection research shows agentic browsers can be trained into phishing traps
Technical Analysis
H score25
First: 11.03.2026 18:38
Last: 11.03.2026 18:38
Sources 1
About this happening:
Perplexity's Comet AI browser is the focus of a technical analysis thread showing how prompt injection and malicious URLs can steer an agentic browser into data...
Perplexity Comet prompt-injection research shows agentic browsers can be trained into phishing traps
Technical AnalysisAbout this happening: Perplexity's Comet AI browser is the focus of a technical analysis thread showing how prompt injection and malicious URLs can steer an agentic browser into data...
Google Looker Studio cross-tenant SQL injection flaws SQL injection flaw
Vulnerability
H score3
First: 10.03.2026 15:20
Last: 10.03.2026 15:20
Sources 1
About this happening:
Researchers disclosed nine cross-tenant vulnerabilities in Google Looker Studio that could let attackers run arbitrary SQL queries on victims' databases and exfiltrate...
Google Looker Studio cross-tenant SQL injection flaws SQL injection flaw
VulnerabilityAbout this happening: Researchers disclosed nine cross-tenant vulnerabilities in Google Looker Studio that could let attackers run arbitrary SQL queries on victims' databases and exfiltrate...
Timeline
-
30.09.2025 13:20 3 articles · 9mo ago
Gemini Trifecta vulnerabilities disclosed and remediated
Initial DisclosureTenable disclosed the "Gemini Trifecta" after finding three flaws in Google Gemini Search Personalization, Gemini Cloud Assist, and the Gemini Browsing Tool that enabled search-injection attacks, prompt injection, and data exfiltration affecting saved information and location data; Google remediated the issues with product-specific fixes, including rolling back the vulnerable Search Personalization model, changing Cloud Assist log-summary hyperlink rendering, and blocking indirect prompt-injection exfiltration in the Browsing Tool.
Show sources
- 'Trifecta' of Google Gemini Flaws Turn AI into Attack Vehicle — www.darkreading.com — 30.09.2025 13:20
- 'Trifecta' of Google Gemini Flaws Turn AI into Attack Vehicle — www.darkreading.com — 30.09.2025 13:20
- Researchers Disclose Google Gemini AI Flaws Allowing Prompt Injection and Cloud Exploits — thehackernews.com — 30.09.2025 16:18