CVE-2020-5953: InsydeH2O UEFI: SMM privilege escalation (not AI-related)

UNKNOWN
Published February 3, 2022
CISO Take

This CVE describes a pointer dereference flaw in the SWSMI handler of InsydeH2O UEFI firmware that lets an attacker with existing kernel (ring 0) access escalate into System Management Mode (ring -2), the most privileged execution context on x86 platforms, effectively below the OS and hypervisor. It was auto-tagged as 'ml_library' due to a name collision between the firmware vendor 'InsydeH2O' and the unrelated AI/ML library 'h2o' — this is a general-purpose firmware vulnerability with no connection to AI/ML software, models, or pipelines. There is no CISA KEV listing, no public exploit or Nuclei template, and the EPSS score of 0.284% places it in a low-probability exploitation band despite requiring local access as a prerequisite anyway. CISOs should route this to firmware/endpoint management teams for standard OEM BIOS patching (Insyde, Siemens, NetApp advisories referenced) rather than to the AI security backlog. No action needed from AI/ML platform or MLSecOps teams specifically.

Sources:NVDEPSS

What is the risk?

Low relevance to AI/ML risk posture. Technically the underlying firmware bug is serious in isolation — SMM code execution provides an attacker with a persistence layer invisible to the OS, EDR, and even hypervisor-based defenses — but it requires local ring-0 access as a precondition, has no assigned CVSS score in this record, no CISA KEV listing, no public PoC, and an EPSS score (0.284%) indicating low likelihood of opportunistic exploitation. This is standard supply-chain/firmware hygiene risk affecting any device running the vulnerable InsydeH2O UEFI build, independent of what software (AI-related or not) runs on top of it.

How does the attack unfold?

Prerequisite kernel access
Attacker already holds ring-0 (kernel) code execution on the target device, typically via a separate OS-level exploit.
SWSMI handler exploitation
Attacker triggers the vulnerable SWSMI handler's GetVariable call, causing an out-of-SMRAM dereference of the gRT pointer.
Privilege escalation to SMM
Execution is redirected into System Management Mode, granting ring -2 privileges above the OS and hypervisor.
Persistent platform compromise
Attacker installs a firmware-level implant that survives OS reinstalls and evades OS/hypervisor-based security tooling, with no AI-specific targeting.

What systems are affected?

PackageEcosystemVulnerable RangePatched
n/a——No patch

Do you use n/a? You're affected.

How severe is it?

CVSS 3.1
N/A
EPSS
0.3%
chance of exploitation in 30 days
Higher than 21% of all CVEs
Exploitation Status
No known exploitation
Sophistication
Advanced

What should I do?

1 step
  1. 1) Identify affected devices via OEM firmware/BIOS version inventory (check against Insyde, Siemens SSA-306654, and NetApp NTAP-20220222-0005 advisories referenced in this CVE). 2) Apply vendor-supplied UEFI firmware updates through standard OEM channels — this is not a package or library CVE and has no application-layer patch. 3) Where firmware updates lag, use platform integrity tooling (e.g., CHIPSEC, TPM-based measured boot/attestation, Secure Boot enforcement) to detect SMM tampering. 4) No AI/ML-specific mitigation is required; treat as standard endpoint/firmware lifecycle management.

How is it classified?

Code ExecutionInference

Which compliance frameworks are affected?

This CVE is relevant to:

NIST AI RMF
MAP 3.4-Third-party/infrastructure risks to AI systems are identified

Frequently Asked Questions

What is CVE-2020-5953?

This CVE describes a pointer dereference flaw in the SWSMI handler of InsydeH2O UEFI firmware that lets an attacker with existing kernel (ring 0) access escalate into System Management Mode (ring -2), the most privileged execution context on x86 platforms, effectively below the OS and hypervisor. It was auto-tagged as 'ml_library' due to a name collision between the firmware vendor 'InsydeH2O' and the unrelated AI/ML library 'h2o' — this is a general-purpose firmware vulnerability with no connection to AI/ML software, models, or pipelines. There is no CISA KEV listing, no public exploit or Nuclei template, and the EPSS score of 0.284% places it in a low-probability exploitation band despite requiring local access as a prerequisite anyway. CISOs should route this to firmware/endpoint management teams for standard OEM BIOS patching (Insyde, Siemens, NetApp advisories referenced) rather than to the AI security backlog. No action needed from AI/ML platform or MLSecOps teams specifically.

Is CVE-2020-5953 actively exploited?

No confirmed active exploitation of CVE-2020-5953 has been reported, but organizations should still patch proactively.

How to fix CVE-2020-5953?

1) Identify affected devices via OEM firmware/BIOS version inventory (check against Insyde, Siemens SSA-306654, and NetApp NTAP-20220222-0005 advisories referenced in this CVE). 2) Apply vendor-supplied UEFI firmware updates through standard OEM channels — this is not a package or library CVE and has no application-layer patch. 3) Where firmware updates lag, use platform integrity tooling (e.g., CHIPSEC, TPM-based measured boot/attestation, Secure Boot enforcement) to detect SMM tampering. 4) No AI/ML-specific mitigation is required; treat as standard endpoint/firmware lifecycle management.

What is the CVSS score for CVE-2020-5953?

No CVSS score has been assigned yet.

What is the AI security impact?

Compliance Controls Affected

NIST AI RMF: MAP 3.4

What are the technical details?

Original Advisory

A vulnerability exists in System Management Interrupt (SWSMI) handler of InsydeH2O UEFI Firmware code located in SWSMI handler that dereferences gRT (EFI_RUNTIME_SERVICES) pointer to call a GetVariable service, which is located outside of SMRAM. This can result in code execution in SMM (escalating privilege from ring 0 to ring -2).

Exploitation Scenario

An attacker who has already achieved kernel-level (ring 0) code execution on a target host — for example via a separate OS or driver vulnerability — triggers the SWSMI handler's flawed GetVariable call path, causing it to dereference the gRT (EFI_RUNTIME_SERVICES) pointer outside of protected SMRAM. This allows the attacker to redirect execution into SMM, gaining ring -2 privileges that persist beneath the OS, hypervisor, and most security tooling. From there, the attacker could implant a firmware-level rootkit that survives OS reinstalls and disk wipes, and could subsequently tamper with any software running on the host — including, incidentally, AI/ML inference services or model artifacts stored on disk — as a downstream consequence of full platform compromise rather than a targeted AI attack.

Timeline

Published
February 3, 2022
Last Modified
August 11, 2026
First Seen
August 11, 2026

Related Vulnerabilities