Cyber Resilience

CVE-2026-3989

Deserialization in Lmsys Sglang ≤ 0.5.10

Public PoCDeserialization
Published
12 March 2026
Modified
10 August 2026
Patch / advisory
CVSS Score v3.1 7.8
Click a component to see what it means
Raw vectorCVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
EPSS Score 0.0037 30th percentile
Risk Priority 56 floored blend · peak EPSS

Summary

CVE-2026-3989 is a high-severity Deserialization of Untrusted Data (CWE-502) vulnerability in Lmsys Sglang. Its CVSS base score is 7.8 (High).

Operationally, exploitation aligns with the MITRE ATT&CK technique Exploit Public-Facing Application (T1190); ranked at the 30th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog; a public proof-of-concept is referenced.

Deeper analysis AI-assisted summary

Synthesised by an AI model from the NVD description and linked references — a reading aid, not an authoritative source.

CVE-2026-3989 is a high-severity vulnerability (CVSS 7.8) in the SGLang framework, specifically affecting the `replay_request_dump.py` script located in the `scripts/playground` directory. The issue stems from an insecure use of `pickle.load()` without proper validation or safe deserialization, allowing arbitrary code execution when processing untrusted pickle files (.pkl). Published on March 12, 2026, this flaw impacts users running the affected script in SGLang, an open-source framework for serving large language models.

An attacker can exploit this vulnerability locally by providing a malicious .pkl file to a victim. Exploitation requires low complexity and user interaction, as the victim must execute the `replay_request_dump.py` script on the crafted file (AV:L/AC:L/PR:N/UI:R). Successful exploitation grants the attacker remote code execution (RCE) on the host device, with high impacts on confidentiality, integrity, and availability (C:H/I:H/A:H), potentially leading to full system compromise.

Mitigation is available through the SGLang project's patch in pull request #20904 and the release of version v0.5.10, which addresses the insecure deserialization. Security advisories, including analysis from Orca Security, highlight this as one of multiple RCE vulnerabilities in the SGLang LLM framework and recommend updating to the patched version while avoiding untrusted .pkl files.

SGLang's role in deploying large language models makes this vulnerability particularly relevant to AI/ML infrastructure, where replay scripts may be used in development or debugging workflows. No public evidence of real-world exploitation has been reported as of the CVE publication.

OWASP Top 10 for Web (2025)

EU & UK References

Vulnerability Data

SGLangs `replay_request_dump.py` contains an insecure pickle.load() without validation and proper deserialization. An attacker can take advantage of this by providing a malicious .pkl file, which will execute the attackers code on the device running the script.

CWE(s)

Related Threats

MITRE ATT&CK Enterprise Techniques

T1190 Exploit Public-Facing Application Initial Access
Adversaries may attempt to exploit a weakness in an Internet-facing host or system to initially access a network.
T1203 Exploitation for Client Execution Execution
Adversaries may exploit software vulnerabilities in client applications to execute code.
T1210 Exploitation of Remote Services Lateral Movement
Adversaries may exploit remote services to gain unauthorized access to internal systems once inside of a network.
Derived from this CVE’s CWE(s) via the direct CWE→ATT&CK cross-walk.

CVEs Like This One

CVE-2026-3060Same product: Lmsys Sglang
CVE-2026-3059Same product: Lmsys Sglang
CVE-2026-7301Same product: Lmsys Sglang
CVE-2026-14890Same product: Lmsys Sglang
CVE-2026-15969Same product: Lmsys Sglang
CVE-2026-7304Same product: Lmsys Sglang
CVE-2026-15976Same product: Lmsys Sglang
CVE-2026-15978Same product: Lmsys Sglang
CVE-2026-7302Same product: Lmsys Sglang
CVE-2026-10775Same product: Lmsys Sglang

Affected Assets

lmsys
sglang
≤ 0.5.10

Mitigating Controls

Likely Mitigating Controls AI

Per-CVE control mapping for this CVE has not run yet; the list below is derived from the weakness types (CWEs) cited in the NVD entry.

addresses: CWE-502

Penetration testing supplies malicious serialized objects, detecting unsafe deserialization and supporting corrective actions.

addresses: CWE-502

Evaluation of untrusted data handling (deserialization testing) reveals unsafe processing, which the required remediation process addresses.

addresses: CWE-502

Untrusted serialized data can be deserialized and observed inside the chamber, blocking gadget-chain exploitation outside the sandbox.

addresses: CWE-502

Validates or rejects untrusted serialized data before deserialization occurs.

addresses: CWE-502

Identifies and blocks malicious code introduced through deserialization of untrusted data at system boundaries.

addresses: CWE-502

Integrity verification of serialized information can detect tampering before deserialization occurs.

addresses: CWE-502

Provenance of associated data allows detection of untrusted sources before deserialization or processing occurs.

Mitigating Controls (NIST CSF 2.0) AI

Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.

PR.PS-02 none match
prevents

PR.PS-02 addresses only post-deployment updates/patching and cannot prevent introduction of unsafe deserialization code, yet it can remediate some instances when the flaw exists in outdated libraries or components.

Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI

Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.

finds

Security testing includes validation of deserialization routines and the use of untrusted data, reducing the likelihood that unsafe object reconstruction will be deployed.

prevents

Requiring vetted libraries, regular updates and SAST before release reduces the likelihood that deserialization logic will accept and act on attacker-controlled serialized objects.

finds

Regular scanning of third-party libraries and timely patching reduce the likelihood that unsafe deserialization vulnerabilities remain active.

none

Mandatory malware scanning of data received over networks or storage media intercepts malicious serialized payloads before they are deserialized by the target application.

References