TaxHacker EUVD-2026-83759

Exploitation requires reaching the testLLMProviderAction server action in app/(app)/apps/settings/actions.ts and controlling its provider/apiKey/model/baseUrl arguments - i.e., access to the TaxHacker settings/LLM-configuration function of a running instance. The core condition is the ability to set an arbitrary LLM provider baseUrl, which the server then contacts. Limiting factors: this is a self-hosted, typically single-tenant/login-gated application, so realistic exploitation needs an authenticated session (contradicting the vendor's PR:N); impact is limited to confidentiality (VC:L) with the SSRF constrained by what internal endpoints the server can reach; and there is no vendor patch, so mitigation depends on network egress controls around the deployment.

This is a moderate-severity SSRF (CWE-918) in a niche open-source, self-hosted application, and the signals broadly agree it is NOT a high-priority issue. The vendor CVSS 4.0 vector scores 5.5 with only VC:L (limited confidentiality) and no integrity or availability impact, reflecting a constrained SSRF rather than an internal-network pivot to RCE. A public POC exists (GitHub issue #187), so the 'E:P' (Proof-of-Concept) exploit-maturity flag in the vector is consistent - publicly available exploit code exists, but this is distinct from active exploitation. There is no CISA KEV listing and no EPSS score in the provided data, and the exploitation status is 'no public exploit weaponized in the wild' beyond the researcher POC. One key discrepancy: the vendor vector states PR:N (unauthenticated), but the vulnerable function testLLMProviderAction lives in an app settings action (app/(app)/apps/settings/actions.ts), which in a self-hosted, login-gated application realistically requires an authenticated session - I dispute PR:N and assess PR:L. Overall this is a real but low-consequence bug in a low-footprint product, further blunted by the vendor being unresponsive (no patch), meaning defenders relying on this software must self-mitigate.

An attacker with access to the TaxHacker settings UI supplies a crafted baseUrl (and provider/apiKey/model arguments) pointing at an internal service or cloud metadata endpoint, then triggers the 'test LLM provider' action so the server issues the request on the attacker's behalf. Because the response or error can leak information, the attacker performs internal reconnaissance or reads limited internal data. A public POC exists in the project's GitHub issue #187, and the attack is remote over the network.