CVE-2026-3503 is a protection-mechanism failure in wolfSSL's wolfCrypt library: on ARM Cortex-M microcontrollers, a physical attacker with fault-injection (glitching) equipment can corrupt seed or pointer values during the Keccak-based expansion step used by the post-quantum ML-KEM and ML-DSA algorithms, potentially recovering key material; NVD has published its own Primary CVSS v3.1 score of 5.2 (medium).
Vulnerability summary
5.2
Medium
0.2%
EPSS score (FIRST)
No
CISA KEV catalog
2026-03-19
Published
Vulnerability ID: CVE-2026-3503 · Source: NVD
Affected components
| Component | CPE | Version range |
|---|---|---|
| wolfSSL (wolfCrypt) | cpe:2.3:a:wolfssl:wolfssl:*:*:*:*:*:*:*:* | 5.8.2 up to (not including) 5.9.0, when running wolfCrypt's ML-KEM/ML-DSA post-quantum code on ARM platforms (Cortex-M microcontrollers); fixed by wolfSSL GitHub pull request #9734 |
What does this mean for affected devices?
CVE-2026-3503 (CWE-335, misuse of a PRNG seed per the CNA's weakness record) is a protection-mechanism failure in wolfSSL's wolfCrypt cryptography library. NVD's description states it affects "wolfCrypt post-quantum implementations (ML-KEM and ML-DSA) in wolfSSL on ARM Cortex-M microcontrollers," allowing "a physical attacker to compromise key material and/or cryptographic outcomes via induced transient faults that corrupt or redirect seed/pointer values during Keccak-based expansion" (NVD).
NVD has completed its own analysis and published a Primary CVSS v3.1 score of 5.2 (medium), vector CVSS:3.1/AV:P/AC:L/PR:N/UI:N/S:U/C:H/I:L/A:N — physical attack vector, low complexity, no privileges or user interaction, high confidentiality impact, low integrity impact, no availability impact. wolfSSL's own CNA record ([email protected]) separately submitted a Secondary CVSS v4.0 score of 4.3 (medium), vector CVSS:4.0/AV:P/AC:H/AT:P/PR:N/UI:N/VC:H/VI:L/VA:N/..., which rates attack complexity higher (physical attack requirements present) than NVD's v3.1 Primary rating; this page's headline score is NVD's own Primary v3.1 figure (NVD).
The flaw is a fault-injection (voltage or clock glitching) attack: it requires an attacker to have physical, hands-on access to the target chip and specialized equipment capable of inducing transient faults during execution — it cannot be triggered remotely over a network. The affected code is wolfSSL's implementation of ML-KEM (post-quantum key encapsulation, formerly CRYSTALS-Kyber) and ML-DSA (post-quantum digital signatures, formerly CRYSTALS-Dilithium) specifically as built for the ARM Cortex-M family of microcontrollers — the dominant MCU core used in IoT sensors, smart meters, building-automation controllers, medical devices, and other embedded products that are now starting to adopt post-quantum cryptography ahead of the EU Cyber Resilience Act and similar mandates. A successful glitch during the Keccak-based seed/pointer expansion step can corrupt or redirect internal state, potentially letting the attacker recover ML-KEM/ML-DSA key material from the device (wolfSSL/wolfssl pull request #9734, "ML-KEM/ML-DSA: harden against fault attacks").
wolfSSL's fix, merged via pull request #9734, hardens the affected routines against fault attacks; the vulnerable code window is wolfSSL 5.8.2 up to (not including) the 5.9.0 release that incorporates the hardening (wolfSSL/wolfssl pull request #9734).
This CVE is not listed in the CISA Known Exploited Vulnerabilities catalog as of catalog version 2026.11.06 (released 2026-11-06), and FIRST scores the probability of exploitation at 0.00162 (5.450th percentile) as of 2026-11-06 — among the lower EPSS scores, consistent with an attack that requires physical possession of the device and specialized fault-injection hardware rather than remote network access (CISA KEV, FIRST EPSS).
Device makers shipping ARM Cortex-M products with wolfSSL's post-quantum ML-KEM/ML-DSA code, particularly in tamper-sensitive or physically-accessible deployments (metering, access control, field sensors), should update to wolfSSL 5.9.0 or later and, where key material value justifies it, pair the software fix with physical tamper-resistance measures.
Evidence
- Protection-mechanism failure (CWE-335) in wolfCrypt's ML-KEM and ML-DSA post-quantum implementations on ARM Cortex-M microcontrollers: a physical attacker can induce transient faults that corrupt seed/pointer values during Keccak-based expansion, compromising key material. — NVD
- NVD Primary CVSS v3.1 score 5.2 (medium), vector CVSS:3.1/AV:P/AC:L/PR:N/UI:N/S:U/C:H/I:L/A:N; wolfSSL's own CNA record separately submitted a Secondary CVSS v4.0 score of 4.3. — NVD
- Affects wolfSSL (wolfCrypt) versions 5.8.2 up to (not including) 5.9.0 on ARM platforms; fixed via pull request #9734 titled "ML-KEM/ML-DSA: harden against fault attacks", authored by SparkiDev. — wolfSSL/wolfssl GitHub pull request #9734
- Not present in the CISA KEV catalog as of catalog version 2026.11.06 (released 2026-11-06). — CISA KEV
- EPSS score is 0.00162 (5.450th percentile) as of 2026-11-06. — FIRST EPSS
Questions about this page?

Related advisories
Sources
Frequently asked questions
Does exploiting this wolfSSL flaw require network access?
No. Per NVD's Primary CVSS v3.1 vector, the attack vector is PHYSICAL — an attacker needs hands-on access to the ARM Cortex-M device and fault-injection (glitching) equipment to corrupt internal state during cryptographic operations; it cannot be triggered remotely over a network.
Which wolfSSL cryptographic operations are affected?
Per NVD and the wolfSSL fix (pull request #9734), the flaw affects wolfCrypt's post-quantum ML-KEM (key encapsulation) and ML-DSA (digital signature) implementations specifically as built for ARM Cortex-M microcontrollers, where induced transient faults can corrupt seed or pointer values during Keccak-based expansion.
Is this already fixed, and in which versions?
Yes. wolfSSL versions 5.8.2 up to but not including 5.9.0 are affected; the hardening fix shipped via wolfSSL/wolfssl pull request #9734, titled "ML-KEM/ML-DSA: harden against fault attacks."
How urgent is this compared to other wolfSSL CVEs disclosed around the same time?
Per FIRST EPSS, CVE-2026-3503 scores 0.00162 (5.450th percentile) as of 2026-11-06 — among the lower exploitation-probability scores, reflecting that the physical, equipment-dependent attack vector is far less scalable than the network-reachable TLS parsing bugs wolfSSL also disclosed in 2026. It is not in the CISA KEV catalog.

