AI Stack Alert Low CVE-2026-123653 For AI Eng, AppSec AI-assisted
CVE-2026-123653: DeepSeek-Reasonix up to 2.52.0 LLM prompt input flaw in discoverDocs
NVD describes CVE-2026-123653 in esengine DeepSeek-Reasonix up to 2.52.0. The flaw is in the discoverDocs function in internal/memory/doc.go and is classed as improper neutralization of input used for LLM prompting. The source says it can be started remotely, a public exploit has been disclosed, and version 2.53.0 (patch 4ca33493a673c2788aa8598fbca2b2f47a69e1b8) addresses it. The CVSS base score is 2.1.
Why it matters
Organizations running DeepSeek-Reasonix up to 2.52.0 are exposed to a remotely triggered input-handling flaw in LLM prompt construction, and public exploit information exists. The source does not describe the resulting impact, and the low CVSS score (2.1) suggests limited severity.
Do this
- Identify all DeepSeek-Reasonix deployments and developer installs at version 2.52.0 or earlier.
- Upgrade affected components to version 2.53.0, which the source says addresses this issue.
- Confirm that commit 4ca33493a673c2788aa8598fbca2b2f47a69e1b8 is included in the deployed build.
- Until upgraded, review which untrusted document sources reach the discoverDocs document discovery path in the memory feature and restrict them where possible.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.21%
Signals
- NoCISA KEV
- 2.1CVSS
- 0.21%EPSS
- NoPublic PoC
- 18/100Priority
- 60%Confidence
Evidence
“A security vulnerability has been detected in esengine DeepSeek-Reasonix up to 2.52.0.”
“The manipulation leads to improper neutralization of input used for llm prompting.”
“Upgrading to version 2.53.0 is able to address this issue.”
Sources
Affected / related
DeepSeek-Reasonix, esengine
Published 2026-11-06 14:17 UTC · updated 2026-11-08 02:26 UTC · AI-assisted summary, checked against quoted sources · How we score