Researchers Trick Perplexity's Comet AI Browser Into Phishing Scam in Under Four Minutes
Summary: Guardio researchers intercepted traffic between Perplexity's Comet AI browser and the vendor's servers and fed it to a Generative Adversarial Network, making the browser fall for a phishing scam in under four minutes. The technique exploits what researcher Shaked Chen calls "Agentic Blabbering" - the browser narrating what it sees, plans, and flags as suspicious - which attackers use as a training signal to iterate a phishing page until the agent stops complaining and enters credentials on a bogus refund page.
What's compelling: Once a scam page is tuned offline against a specific AI browser, it works on every user running that same agent, shifting the target from the human to the model. The disclosure follows Trail of Bits' four prompt injection techniques against Comet and Zenity Labs' two zero-click PerplexedComet attacks, and OpenAI's December 2025 statement that such weaknesses are "unlikely to ever" be fully resolved in agentic browsers.
Read the original at The Hacker News ↗
- Topics
- AI Security
- Related
- SECURITY AFFAIRS AI-CYBERSECURITY NEWSLETTER ROUND 8 Cyber Risks · Nov 15, 2026Pentagon Renames CDAO ‘Super Intelligence Office’ Cyber Risks · Nov 15, 2026186 Fake AI Domains Redirect Users to Click Fix Lures, Browser Malware, and Search Hijackers Cyber Risks · Nov 12, 2026