Company Research Agent EUVD-2026-97001

Exploitation requires that the application's /generate-pdf endpoint be network-reachable and accept attacker-controlled report_content (the default behavior in Company Research Agent through 2.2.0, with no authentication per PR:N). The specific trigger is injecting unescaped ReportLab paragraph markup - inline <img> elements - into report_content, which ReportLab resolves by fetching the supplied src URL server-side. Limiting factors: impact is constrained to outbound request forgery and leakage of responses that can be rendered as images into the returned PDF (confidentiality-low, no integrity or availability impact), so non-image or non-renderable internal responses yield mainly reachability/probing signal rather than full content exfiltration; realistic internal-network risk also depends on whether the instance is internet-exposed and what internal services or metadata endpoints are routable from the server.

Signals converge on a genuine-but-moderate SSRF rather than a top-tier emergency. The vendor CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:N/UI:N) confirms a network-reachable, low-complexity, unauthenticated flaw with no user interaction, which supports easy triggering; however impact is deliberately scoped low (VC:L/VI:N/VA:N, SC:L) because the primary outcome is outbound request forgery and leakage of image-renderable responses, not direct code execution or full data compromise, yielding a Medium 6.9. No EPSS score was provided, so exploitation probability cannot be quantified. The CVE is NOT in CISA KEV and no KEV-style active exploitation is claimed. A detailed technical writeup (hackmd) and a VulnCheck advisory with direct source-line references exist, so exploitation methodology is public even though no packaged exploit flag was supplied in the input. There are no conflicting signals between the vector and the description - both agree on remote, unauthenticated, low-impact SSRF. The main risk amplifier is deployment context: if an instance is reachable from the internet while sitting inside a network with sensitive internal services or cloud metadata endpoints, the low technical impact understates the real blast radius. Overall this is a real but second-tier priority, driven by ease of exploitation and internal-recon potential rather than by severity or confirmed in-the-wild use.

An unauthenticated attacker sends a request to the /generate-pdf endpoint with report_content containing a ReportLab paragraph <img> tag whose src points to an internal host or cloud metadata address. The server renders the PDF, fetching the attacker-chosen URL and embedding the image response, which lets the attacker probe internal reachability and exfiltrate image-renderable content via the returned PDF. The technique is publicly documented in the referenced hackmd writeup and VulnCheck advisory, though no weaponized exploit is cited in the input.