Severity

High · CVSS 7.5

Target

Major AI Platform

Date

2026-02

Category

Privacy / Information Disclosure

Mass User Prompt Leakage via URL Parameters: Permanently Archived

The AI platform accepted user prompts via URL query parameters, which were reflected in the page HTML. These prompt-carrying URLs were actively crawled and permanently archived by major web archival services. Prompts contained source code with business logic, system prompt instructions, and corporate intelligence.

Attack Chain

Step 1: Prompts accepted via URL parameters

URL query parameters accepted prompt content and reflected it in the page source.

Step 2: Web archives actively capturing prompts

Over a thousand prompt-carrying URLs permanently archived and publicly searchable.

Step 3: Sensitive data categories in cached prompts

Prompts contained source code with financial logic, AI system prompts, messaging API data, and corporate intelligence.

Step 4: No archival prevention in place

The page had Cache-Control headers but no archival-prevention directives. Web crawlers ignore Cache-Control for archival decisions.

What Scanners Would Miss

Web scanners test for XSS in reflected parameters. They don't cross-reference web archives to discover that URL parameters are creating a permanent, searchable database of user prompts.

By the Pinaka team · Published for educational purposes. All findings were responsibly disclosed.