Brave Uncovers Vulnerability in Perplexity Comet's AI Assistant, Enabling Malicious Prompt Injection

Brave uncovers a vulnerability in Perplexity Comet's AI assistant that enables malicious prompt injection, allowing attackers to manipulate it into performing unauthorized actions like data exfiltration; mitigations include separating user instructions, checking alignment, requiring interaction for sensitive tasks, and isolating agentic browsing.

• Brave discovers a vulnerability in Perplexity Comet's agentic browsing implementation that allows attackers to inject malicious prompts into webpage content
• The vulnerability enables attackers to manipulate Comet's AI assistant into performing unauthorized actions like accessing user accounts and exfiltrating sensitive data
• Brave outlines potential mitigations including separating user instructions from webpage content, checking user-alignment for tasks, requiring user interaction for sensitive actions, and isolating agentic browsing