CometJacking Attack Hijacks Perplexity AI Browser to Steal User Data

Cybersecurity researchers have uncovered a new attack called CometJacking, which targets Perplexity's AI-powered browser, Comet. The attack leverages malicious prompts hidden within seemingly harmless links to siphon sensitive data, including emails, calendar entries, and user memory. This exploit bypasses Perplexity's security measures using Base64-encoding tricks, turning the AI browser into an unwitting data thief.

How CometJacking Works

The attack unfolds in five steps:

1. A victim clicks on a specially crafted URL, delivered via phishing or embedded in a webpage.
2. Instead of navigating to the intended destination, the URL triggers Comet's AI to execute a hidden prompt.
3. The AI captures data from connected services like Gmail and Calendar.
4. The stolen data is obfuscated using Base64 encoding.
5. The encoded payload is sent to an attacker-controlled endpoint.

Michelle Levy, Head of Security Research at LayerX, warned, "This isn't just about stealing data; it's about hijacking the agent that already has the keys." The attack exploits Comet's trusted status, as it already has authorized access to user accounts.

Perplexity's Response and Broader Implications

Perplexity has dismissed the findings as having "no security impact," but experts argue that AI-native tools introduce new risks. Earlier this year, Guardio Labs revealed Scamlexity, a similar attack tricking AI browsers into interacting with phishing pages.

Or Eshed, CEO of LayerX, emphasized, "AI browsers are the next enterprise battleground." Organizations are urged to implement controls to detect and neutralize malicious prompts before widespread exploitation occurs.

For more details on the research, visit LayerX's report.

Follow The Hacker News for updates on emerging cybersecurity threats.