CVE-2026-22778Disclosure(vllm / vllm)

LOWCVSS 9.8 · CRITICAL

Exploit discussion active in current signal (1 latest mentions)

Immediate actions

Patch vllm vllm systems immediately
Hunt for exploitation attempts and persistence artifacts
Increase monitoring for publicly documented tradecraft

Recommended action window: High priority (within 72h)

NVD description

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.

vllm CVEs Back to dashboard

2.3/ 10 priority

Sources & remediation

Patch / fix

Vendor / third-party advisories

https://github.com/vllm-project/vllm/security/advisories/GHSA-4r2x-xpjr-7cvv

Other references

Weakness type (CWE)

CWE-532CWE-209

Priority

LOW

Exploitation

NONE

PoC

YES

Patch

AVAILABLE

Momentum

STABLE

Are you affected?

If you run products in this scope, you should treat this CVE as relevant to your environment.

vllm

Threat summary

Public PoC is present in monitored signal
Patch or workaround signal is available
22 mentions across 12 observed days
Momentum state: stable

What's happening

PoC mentioned or linked in 4 signals
Patch or workaround mentioned in 6 signals
Technical details provided in 16 signals
Disclosure: 11 classified signals
General: 4 classified signals
Peaked 10d ago at 6 mentions (2026-02-03); latest day: 1
22 total mentions across 12 days

Affected systems

Vendors

vllm

Products

vllm

Deep dive

Activity timeline22 mentions / 12d

Signal classification4 categories

Disclosure

1150.0%

Patch

627.3%

General

418.2%

PoC

14.5%

Referenced assets17 URLs

By indicator

Classification over time

Date	Total	Labels

2026-02-02	3	Disclosure2Patch1
2026-02-03	6	Disclosure2General1Patch3
2026-02-04	1	Patch1
2026-02-05	1	Disclosure1
2026-02-06	2	Disclosure2
2026-02-08	1	Disclosure1
2026-02-13	1	Disclosure1
2026-03-11	1	Disclosure1
2026-04-08	1	Patch1
2026-06-05	2	General1PoC1
2026-06-22	2	General2
2026-06-28	1	Disclosure1

Full discourse20 posts

ZoomEye@zoomeye_team244d
Disclosure
🚨 CVE-2026-22778 (CVSS N/A): vLLM Vulnerable to Remote Code Execution vLLM is vulnerable to unauthorized model access via improper authentication in the OpenAI-compatible API server, allowing unauthenticated attackers to list, load, or unload models and exfiltrate sensitive model weights. Search by vul.cve Filter 👉 vul.cve="CVE-2026-22778" ZoomEye Dork 👉 app="vLLM" 294 exposed instances. 🔥 CVE Feed Intelligence: https://www.darkeye.org/vuln/cve/CVE-2026-22778 ZoomEye Link: https://www.zoomeye.ai/searchResult?q=dnVsLmN2ZT0iQ1ZFLTIwMjYtMjI3Nzgi&utm_source=twitter&utm_medium=social&utm_campaign=cve_ops_20260206 Refer: https://nvd.nist.gov/vuln/detail/CVE-2026-22778 #ZoomEye #NetSec #OSINT #CyberSecurity #vLLM #AIsecurity #LLMsecurity #ZeroDay
Post summary
The tweet announces CVE-2026-22778, detailing RCE in vLLM due to improper authentication and noting 294 exposed instances, but provides no PoC, exploit, or patch information.
04027122.9K
11.9K followersView on X
pdnuclei-bot@pdnuclei_bot102d
Disclosure
🚨 CVE-2026-22778 - critical 🚨 vLLM 0.8.3 - 0.14.0 - Information Disclosure > vLLM 0.8.3 to - 0.14.1 contains an information disclosure caused by leaking a heap ad... 👾 https://cloud.projectdiscovery.io/library/CVE-2026-22778 @pdnuclei #NucleiTemplates #cve
Post summary
The tweet alerts about a critical information‑disclosure vulnerability in selected vLLM versions, points to a library resource, but does not provide PoC or active exploitation details.
00021372
1.3K followersView on X
CVE@CVEnew107d
General
CVE-2026-54236 vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a sanitize_message helper t… https://www.cve.org/CVERecord?id=CVE-2026-54236
Post summary
The text merely notes the existence of CVE-2026-54236 related to vLLM, without any detailed vulnerability, exploit, or patch information.
01011785
57.7K followersView on X
Cybersecurity News Everyday@TweetThreatNews246d
Patch
A critical vulnerability CVE-2026-22778 in vLLM enables remote code execution via malicious video URLs on multimodal AI endpoints. Affected versions: >=0.8.3 & <0.14.1; patch vLLM 0.14.1 addresses this. #AIFlaw #RemoteExec #Python https://ift.tt/5nbXtSW
Post summary
A critical RCE flaw in vLLM has been disclosed with a patch released (0.14.1); no PoC or active exploitation is reported.
01010106
3.6K followersView on X
Fernando Karl@fernandokarl247d
Patch
🚨 Atenção, profissionais de #cibersegurança! A vulnerabilidade CVE-2026-22778 no vLLM permite execução remota de código via URL de vídeo maliciosa. Atualize para vLLM 0.14.1 IMEDIATAMENTE para proteger seus dados! 🔒 Saiba mais: https://orca.security/resources/blog/cve-2026-22778-vllm-rce-vulnerability/ #RCE #vLLM
Post summary
The post announces CVE-2026-22778, a remote code execution flaw in vLLM triggered by malicious video URLs, and urges users to immediately upgrade to vLLM 0.14.1 to mitigate the risk.
0101094
253 followersView on X
Eli Woodward@ElijahWoodward9247d
General
Full article https://www.ox.security/blog/cve-2026-22778-vllm-rce-vulnerability/
Post summary
The text merely references a blog article URL about CVE-2026-22778 without providing further details.
0101084
226 followersView on X
CVEFind.com@CveFindCom247d
Disclosure
[CVE-2026-22778: CRITICAL] In vLLM versions 0.8.3 to 0.14.0, sending an invalid image resulted in a PIL error leak, exposing a heap address, making ASLR easier to bypass. Pairing this exploit with a heap ove...#cve,CVE-2026-22778,#cybersecurity https://cvefind.com/CVE-2026-22778
Post summary
The tweet discloses a critical CVE in vLLM that leaks heap addresses via a PIL error, but provides no PoC, exploit code, patch, or evidence of active exploitation.
01010114
583 followersView on X
SRG@SimeonGarratt125d
General
@PsudoMike The satirical part is - their core infra/backend is already completely vulnerable and exposed. And it hasn't even started. (CVE-2026-22778), (CVE-2026-24779)... for starters.
Post summary
The tweet highlights the discovery of vulnerabilities (CVE-2026-22778, CVE-2026-24779) but offers no evidence of exploitation, remediation, or detailed technical info, making it a general disclosure.
10000772
1.3K followersView on X
Firmis Labs@FirmisLabs183d
Patch
CVE-2026-22778 · NIST 9.8/10 https://nvd.nist.gov/vuln/detail/CVE-2026-22778 ask your AI: "check if my project uses vllm and if it's below version 0.14.1" then: "update vllm to the latest version and make sure video processing still works"
Post summary
The message references CVE-2026-22778 with a high CVSS score and advises updating vllm to mitigate the vulnerability.
1000032
1 followersView on X
DotTechES@DotTechES237d
Disclosure
Ojo si usáis vLLM para inferencia local. Se confirma CVE-2026-22778 (9.8/10). Un simple vídeo malicioso te da RCE. Y OpenClaw llenándose de malware. La "Era de los Sistemas" va de asegurar la cadena de suministro, no de picar código rápido.
Post summary
The post confirms the high‑severity CVE‑2026‑22778, highlighting a malicious video that achieves RCE, but no active exploitation or patch information is provided.
1000060
1.0K followersView on X
Brandon Veiseh@BVeiseh246d
Disclosure
If you are running vLLM with video models, make sure to bump your versions. A pretty nasty security vulnerability was discovered yesterday. https://orca.security/resources/blog/cve-2026-22778-vllm-rce-vulnerability/?utm_source=linkedin&utm_medium=organic+social&utm_campaign=orca+blog
Post summary
A new CVE (CVE-2026-22778) affecting vLLM with video models has been discovered, prompting a recommendation to update versions.
0001088
378 followersView on X
Moshe Siman Tov Bustan@MosheTov247d
Patch
🚨 RCE ON vLLM! PATCH NOW! 🚨 A recently discovered vulnerability (CVE-2026-22778) in vLLM allows threat actors to send a malicious link to a vLLM service with the “video model” enabled in order to trigger an RCE, allowing complete takeover of the server! https://www.ox.security/blog/cve-2026-22778-vllm-rce-vulnerability/
Post summary
The text announces a remote code execution vulnerability in vLLM (CVE‑2026‑22778) and urges users to apply the available patch immediately.
01000149
78 followersView on X
Infoflowcloud@infoflowcloud107d
General
🚨*CVE* CVE-2026-54236 vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a sanitize_message helper t… https://www.cve.org/CVERecord?id=CVE-2026-54236 ----- Traducción: CVE-2026-54236 vLL… http://infoflow.cloud`
Post summary
The text merely references CVE‑2026‑54236, providing a link to the CVE record without further details or actionable information.
0000033
88 followersView on X
SRG@SimeonGarratt125d
PoC
@RealNicoLagan Bumping this b/c 🙏 "Canada's 'big plan' is sitting with core infra/backend already completely vulnerable and exposed. If the govt is serious [PoC] (CVE-2026-22778), (CVE-2026-24779)... for starters."
Post summary
The tweet references CVE‑2026‑22778 and CVE‑2026‑24779, indicating that a proof‑of‑concept exists for these vulnerabilities affecting Canadian core infrastructure.
0000066
1.3K followersView on X
RagingCISO@CisoRaging77913211d
Disclosure
CVE-2026-22778: vLLM ASLR bypass → RCE. Invalid image triggers error that leaks heap addresses to client. Chain with existing overflow = silent RCE. Your cutting-edge LLM engine falls to an ancient PIL bug. AI stack maturity: pathetic.
Post summary
CVE-2026-22778 describes an ASLR bypass that leads to a silent remote code execution via an invalid image error leaking heap addresses and chaining with an existing overflow. No PoC, exploit code, patch, or evidence of active exploitation is mentioned.
0000032
5 followersView on X
The AI generalist@AIengineerlife242d
Disclosure
🚨 CVE-2026-22778 (CVSS 9.8): Critical RCE in vLLM AI library allows unauthenticated attackers to execute code via malicious video URLs. AI infrastructure under attack! Track 777+ CVEs with real-time alerts at https://threatmonitor.io #cybersecurity #CVE #infosec #vulnerability
Post summary
A critical RCE vulnerability (CVE‑2026‑22778) in the vLLM AI library is disclosed, allowing unauthenticated code execution via malicious video URLs; no patch or exploitation evidence is mentioned.
0000057
7 followersView on X
The AI generalist@AIengineerlife244d
Disclosure
🚨 CVE-2026-22778: Critical RCE in vLLM (CVSS 9.8) allows unauthenticated attackers to execute code via crafted video URLs. Track this & 777+ CVEs at https://threatmonitor.io Stay ahead of threats with real-time alerts! #cybersecurity #CVE #infosec #vulnerability
Post summary
The tweet announces CVE‑2026‑22778, a critical remote code execution flaw in vLLM that permits unauthenticated attackers to run code via crafted video URLs, but provides no evidence of exploitation, patches, or PoC.
0000056
7 followersView on X
Eyal Estrin ☁️@eyalestrin245d
Disclosure
Critical RCE in vLLM Allows Server Takeover via Malicious Video URL (CVE-2026-22778) http://dlvr.it/TQmWS0 #appsec
Post summary
A critical remote code execution flaw (CVE-2026-22778) in vLLM permits server takeover through a malicious video URL; the post offers no PoC, exploit code, or patch details.
0000082
2.0K followersView on X
PulsePatch.io@pulsepatchio247d
Patch
vLLM has a Remote Code Execution vulnerability (CVE-2026-22778) in video processing. Update to 0.14.1 or later. #vLLM #RCE #infosec https://www.pulsepatch.io/posts/cve-2026-22778-vllm-rce-video-processing
Post summary
vLLM CVE-2026-22778 is a remote code execution flaw in video processing; users are advised to update to v0.14.1 or newer to mitigate the risk.
0000044
1 followersView on X
Cyber Daily News@CyberDaily_News247d
Patch
vLLM 0.8.3-0.14.0 lets an attacker send a crafted video URL, leak a heap address and trigger a JPEG2000 overflow for full RCE (CVE-2026-22778). Patch to 0.14.1 or disable multimodal endpoints now. https://thecyberexpress.com/cve-2026-22778-vllm-rce-malicious-video-link/ #infosec #CVE2026-22778 #vLLM #RCE #AIsec
Post summary
vLLM 0.8.3-0.14.0 is vulnerable to full RCE via a JPEG2000 overflow triggered by crafted video URLs; patching to 0.14.1 or disabling multimodal endpoints is recommended, and a PoC link is provided.
0000067
12 followersView on X

CPE platform detail1 entries

Part:

Vendor:

Product:

Version:

Target SW:

Target HW:

1 of 1 entries

Part	Vendor	Product	Version	Target SW	Target HW
App	vllm	vllm	-	-	-

Explore more

Back to dashboard vllm CVEs vllm vulnerabilities

What researchers are saying

CVE-2026-22778

ZoomEye[verified]@zoomeye_teamDisclosure

The tweet announces CVE-2026-22778, detailing RCE in vLLM due to improper authentication and noting 294 exposed instances, but provides no PoC, exploit, or patch information.

Eli Woodward[verified]@ElijahWoodward9General

The text merely references a blog article URL about CVE-2026-22778 without providing further details.

SRG[verified]@SimeonGarrattGeneral

The tweet highlights the discovery of vulnerabilities (CVE-2026-22778, CVE-2026-24779) but offers no evidence of exploitation, remediation, or detailed technical info, making it a general disclosure.

Firmis Labs[verified]@FirmisLabsPatch

The message references CVE-2026-22778 with a high CVSS score and advises updating vllm to mitigate the vulnerability.

Brandon Veiseh[verified]@BVeisehDisclosure

A new CVE (CVE-2026-22778) affecting vLLM with video models has been discovered, prompting a recommendation to update versions.

SRG[verified]@SimeonGarrattPoC

The tweet references CVE‑2026‑22778 and CVE‑2026‑24779, indicating that a proof‑of‑concept exists for these vulnerabilities affecting Canadian core infrastructure.

The AI generalist[verified]@AIengineerlifeDisclosure

A critical RCE vulnerability (CVE‑2026‑22778) in the vLLM AI library is disclosed, allowing unauthenticated code execution via malicious video URLs; no patch or exploitation evidence is mentioned.

The AI generalist[verified]@AIengineerlifeDisclosure

The tweet announces CVE‑2026‑22778, a critical remote code execution flaw in vLLM that permits unauthenticated attackers to run code via crafted video URLs, but provides no evidence of exploitation, patches, or PoC.