CVE-2026-22778

🎯 CVE-2026-22778 🧪 PoC 🇪🇺 EUVD

💬 0 Diskussion 📄 .md Alle CVEs anzeigen ✕

CVE-2026-22778 | vLLM is an inference and serving engine for large language models (LLMs). From 0.8.3 to before 0.14.1, when an invalid image is sent to vLLM's multimodal endpoint, PIL throws an error. vLLM returns this error to the client, leaking a heap address. With this leak, we reduce ASLR from 4 billion guesses to ~8 guesses. This vulnerability can be chained a heap overflow with JPEG2000 decoder in OpenCV/FFmpeg to achieve remote code execution. This vulnerability is f

vLLM is an inference and serving engine for large language models (LLMs). From 0.8.3 to before 0.14.1, when an invalid image is sent to vLLM's multimodal endpoint, PIL throws an error. vLLM returns this error to the client, leaking a heap address. With this leak, we reduce ASLR from 4 billion guesses to ~8 guesses. This vulnerability can be chained a heap overflow with JPEG2000 decoder in OpenCV/FFmpeg to achieve remote code execution. This vulnerability is fixed in 0.14.1.

Klassifikation & Betroffenheit:

CWE-209CWE-532 vllm-project vllm 0.8.3, < 0.14.1vllm vllm *

Generation of Error Message Containing Sensitive Information 🎯 High

The product generates an error message that includes sensitive information about its environment, users, or associated data.

🛡️ Empfohlene Mitigation: Ensure that error messages only contain minimal details that are useful to the intended audience and no one else. The messages need to strike the balance between being too cryptic (which can confuse users) or being too detailed (which may reveal more than intended). The messages should not reveal the methods that were …

Vollständige Definition bei MITRE ➔

⚔️ Typische Angriffsmuster (CAPEC):

CAPEC-7 · Blind SQL Injection (High) CAPEC-54 · Query System for Information (Low) CAPEC-215 · Fuzzing for application mapping (Low) CAPEC-463 · Padding Oracle Crypto Attack (High)

🧩 Ähnliche Schwachstellen (Hersteller/Klasse/Score-Band):

CVE-2025-32444 10.0 CVE-2025-46658 9.8 CVE-2025-47277 9.8 CVE-2024-11041 9.8 CVE-2024-9053 9.8

🧪 Exploit-Evidenz:

🐙 joaovicdev/EXPLOIT-CVE-2026-2277

🩹 Patch verfügbar (OSV):

🩹 d7de043d55d1dd629554467e23874097e1c48993 (Commit)

🇩🇪 BSI-Sicherheitshinweise: BSI · hoch · vllm: Schwachstelle ermöglicht Codeausführung ↗

📚 Referenzen & Quellen:

🇪🇺 Original-Eintrag (ENISA EUVD) 🇺🇸 NVD-Eintrag

Ausnutzungs-Zeitleiste:

Reserviert 2026

Veröffentlicht 02/2026

Heute 28.10.

CVSS-Vektor-Analyse: 9.8

AV · Angriffsvektor Netzwerk

AC · Komplexität Gering

PR · Privilegien Keine

UI · Interaktion Keine

S · Scope Unverändert

C · Vertraulichkeit Hoch

I · Integrität Hoch

A · Verfügbarkeit Hoch

CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

CNA-Bewertung (GitHub_M): 9.8 CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

Veröffentlicht:02.02.2026

Aktualisiert:15.07.2026 02:18

Assigner (CNA):GitHub_M

EUVD-ID:EUVD-2026-5277

Quellen: 🇪🇺 EUVD-Datenbank (ENISA) + 🇺🇸 NVD-Anreicherung · 24-h-Cache

CWE-209 ✓ Offizieller Patch / Advisory verfügbar

💡 Gegenmaßnahme: Sicherheits-Update des Herstellers zeitnah einspielen und Netzwerksegmentierung prüfen.

Tier	CVEs	Anteil
≥90 %	740	0,2 %
≥50 %	3.215	0,8 %
≥10 %	22.034	5,8 %
<10 %	354.740	93,2 %

9.8 CRITICAL

EPSS

18.5%

CVE-2026-22778 🌐 Netzwerk (Remote) 🔓 Keine Authentifizierung nötig

vllm-project

CVE-2026-22778 | vLLM is an inference and serving engine for large language models (LLMs). From 0.8.3 to before 0.14.1, when an invalid image is sent to vLLM's multimodal endpoint, PIL throws an error. vLLM returns this error to the client, leaking a heap address. With this leak, we reduce ASLR from 4 billion guesses to ~8 guesses. This vulnerability can be chained a heap overflow with JPEG2000 decoder in OpenCV/FFmpeg to achieve remote code execution. This vulnerability is f

CWE-209 ✓ Offizieller Patch / Advisory verfügbar

💡 Gegenmaßnahme: Sicherheits-Update des Herstellers zeitnah einspielen und Netzwerksegmentierung prüfen.

vor 8 Monaten ↻ vor 3 Monaten

🛡️ Details ➔

Zero-Day & Vulnerability Intelligence Hub