What happened
Company Research Agent renders user-supplied report_content into PDFs through ReportLab without escaping markup. Inline img elements cause the server to make outbound fetches to arbitrary hosts; image responses are embedded and visible in the returned PDF, and reachability of internal services is probed — all without authentication.
Why it matters
An AI research-agent product whose PDF export endpoint doubles as an unauthenticated SSRF lets outsiders reach internal-only services and read image-based responses, pivoting from the agent UI into the host's internal network.
Attack vector
Unauthenticated attacker injects unescaped ReportLab paragraph markup (inline img elements) into report_content of /generate-pdf; the server's PDF renderer fetches the target URL, leaking returned images into PDFs and probing internal reachability.
Affected systems
company-research-agent <= 2.2.0 (/generate-pdf endpoint)
Mitigation
No released fix at publication; sanitize report_content markup and add authentication to /generate-pdf (VulnCheck advisory).