Using threat modeling and prompt injection to audit Comet
blog.trailofbits.com | blog | #ai-security | #browser-security | #prompt-injection | #security-audit | #ai-agents | #threat-modeling | #gmail | #trail-of-bits | #perplexity | #comet
Summary
Trail of Bits used TRAIL threat modeling and adversarial testing to demonstrate four prompt injection techniques against Perplexity's Comet AI that could exfiltrate private Gmail data.
- Published
- Collected
--- id: 1297 title: "Using threat modeling and prompt injection to audit Comet" source_domain: blog.trailofbits.com source_url: "https://blog.trailofbits.com/2026/02/20/using-threat-modeling-and-prompt-injection-to-audit-comet/" published_at: "2026-02-20 16:00" tags: ["ai-security", "browser-security", "prompt-injection", "security-audit", "ai-agents", "threat-modeling", "gmail", "trail-of-bits", "perplexity", "comet"] phying_url: "https://phying.com/item/1297" lang: en --- # Using threat modeling and prompt injection to audit Comet ## Summary Trail of Bits used TRAIL threat modeling and adversarial testing to demonstrate four prompt injection techniques against Perplexity's Comet AI that could exfiltrate private Gmail data. ## Intelligence Metadata - **Sources:** blog.trailofbits.com - **Published:** 2026-02-20 16:00 - **Original:** https://blog.trailofbits.com/2026/02/20/using-threat-modeling-and-prompt-injection-to-audit-comet/ - **Phying News:** https://phying.com/item/1297