CVE-2026-111392
Deferred Deferred - Pending Action

SGLang Multimodal Cache Poisoning via mm_hashes

Publication date: 2026-10-16

Last updated on: 2026-10-16

Assigner: VulnCheck

Description
SGLang through 0.5.21 trusts caller-supplied mm_hashes values as the multimodal prefix-cache key, allowing remote attackers to poison cached KV entries across requests. Attackers can send a different image under a known or predicted hash via /generate to disclose another request's image content or corrupt victims' responses.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published
2026-10-16
Last Modified
2026-10-16
Generated
2026-10-19
AI Q&A
2026-10-16
EPSS Evaluated
2026-10-18
NVD
CVE-2026-111392
EUVD
EUVD-2026-103024

Affected Vendors & Products

Showing 1 associated CPE
Vendor Product Version / Range
sgl-project sglang 0.5.13

Helpful Resources

Exploitability

CWE
CWE Icon
KEV
KEV Icon
CWE ID Description
CWE-345 The product does not sufficiently verify the origin or authenticity of data, in a way that causes it to accept invalid data.

Attack-Flow Graph

AI Quick Actions

Instant insights powered by AI
Executive Summary

This vulnerability in SGLang through 0.5.21 allows attackers to manipulate cached key-value entries by sending an image with a known or predicted hash via the /generate endpoint. The system uses the client-provided hash as the cache key instead of computing it from the actual image content, enabling cache poisoning and cross-tenant image disclosure.

Detection Guidance

Monitor network traffic for unusual /generate endpoint requests containing mm_hashes parameters. Check SGLang logs for mismatches between declared hashes and actual image content. Use tools like Wireshark to inspect HTTP requests for suspicious payloads.

Impact Analysis

Attackers could access cached image content from other users' requests if they reuse a known hash. This could lead to unauthorized data disclosure, corrupted responses, or service disruptions by forcing the system to serve incorrect cached data based on manipulated hashes.

Compliance Impact

This vulnerability could lead to unauthorized access to sensitive data, violating GDPR's data protection principles and HIPAA's confidentiality requirements. It may result in data breaches, unauthorized disclosures, and non-compliance with privacy regulations due to insufficient data authenticity verification.

Mitigation Strategies

Upgrade SGLang to a patched version. Disable the /generate endpoint if unused. Implement network-level controls to block untrusted mm_hashes parameters. Enable the --trust-caller-mm-hashes flag only if required for external KV routers.

Chat Assistant

Ask questions about this CVE
Hi! I’m here to help you understand CVE-2026-111392. Ask me anything about the vulnerability, its impact, or mitigation strategies.
0/70

EPSS Chart