CVE-2026-73560: vLLM multimodal processor vulnerable to arbitrary file reads and SSRF

# CVE-2026-73560: vLLM multimodal processor vulnerable to arbitrary file reads and SSRF

https://gcdradar.vercel.app/incidents/cve-2026-73560-vllm-multimodal-processor-vulnerable-to-arbit · AIR-2026-0086 · closed 2026-08-17

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2026-08-17
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2026-73560
- CVSS: 6.5

## Summary

A vulnerability in the vLLM multimodal processor prior to version 0.26.0 bypasses media domain restrictions, permitting unauthorized file reads and server-side requests. Version 0.26.0 resolves this vulnerability.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-09-08: CVE-2026-73560: vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections (https://github.com/advisories/GHSA-4hhp-h66f-j5j7)
- 2026-08-17: CVE-2026-73560: vLLM is an inference and serving engine for large language models. (https://nvd.nist.gov/vuln/detail/CVE-2026-73560)

## Sources

- 2026-08-17 NVD: CVE-2026-73560: vLLM is an inference and serving engine for large language models. https://nvd.nist.gov/vuln/detail/CVE-2026-73560
- 2026-09-08 GitHub Advisory Database: CVE-2026-73560: vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections https://github.com/advisories/GHSA-4hhp-h66f-j5j7

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.