Security Vulnerability: Information Disclosure via GET Parameter in Perplexity AI GPT-4

Author: Ragavendra Krishna Kumar

Severity: 🚨 MEDIUM

🏷 Vendor

Perplexity AI (https://www.perplexity.ai)

🖥 Product

Perplexity AI Web Application

🆚 Versions Affected

2.51.0 (GPT-4)

🛠️ Vulnerable Component

Shared Chat Session URLs (/search/<token> endpoint)

📌 Summary

An issue in Perplexity AI GPT-4 allows a remote attacker to obtain sensitive information via a GET parameter without proper authorization.

---

⚠️ Impact

This vulnerability enables unauthorized access to sensitive chatbot content, including:

• Legal, medical, or financial questions
• Personally Identifiable Information (PII)
• Corporate or proprietary information

Major Risks:

• Severe privacy breach, especially if shared unknowingly
• Regulatory compliance risks, including:
  • GDPR (Article 5, 32): Violations of confidentiality and data minimization
  • CCPA: Unauthorized access to user data

🔗 Affected Endpoint

https://www.perplexity.ai/search/

Example token: hi-RxWG5knCTUurDLABy27PMg

---

Steps to Reproduce

1. Log in to your Perplexity AI account (or proceed as a guest).
2. Obtain or observe another user's chat token (e.g., via shared link or recon).
3. Paste a modified URL such as: https://www.perplexity.ai/search/hi-RxWG5knCTUurDLABy27PMg
4. Open the link in incognito mode (even without being logged in).
5. Observe that the chat history of another user is accessible without authentication.

---

🧪 Proof of Concept (PoC)

• Performed successfully in incognito mode.
• No login required to retrieve another user's chat via token.
• Confirmed repeatable with multiple valid tokens.
• The link is accessed by someone else (via referrer leak, history, logs, etc.).
• The full conversation is visible — no authentication required.