CVE-2026-42010: vLLM NUL-byte username bypasses

CISO Take

A flaw in gnutls causes RSA-PSK TLS servers to wrongly match a username containing an embedded NUL character against a different, truncated identity, letting an attacker with a valid low-privilege PSK credential authenticate as someone else and bypass intended access controls. This matters because gnutls sits underneath 129 tracked downstream dependents and has racked up 78 other CVEs, and Red Hat's own advisories confirm the bug reaches AI inference infrastructure directly — the RHEL AI Inference Server images (model-opt-cuda-rhel9, vllm-rocm-rhel9) and the discovery-server/discovery-ui components ship the affected library. There is no EPSS score, no CISA KEV listing, no public exploit code, and no Nuclei template as of this writing, so this looks like a quiet library-level bug rather than an actively weaponized one — but CVSS 7.1 (network, low complexity, no user interaction, high confidentiality impact) means it's cheap to exploit once someone builds a proof of concept. Patch gnutls/libtasn1 to the versions fixed in the Red Hat errata (RHSA-2026:13274 and related), rebuild any RHAIIS/vllm-rocm-rhel9 container images from the patched base, and where RSA-PSK is used to authenticate between AI inference nodes or gateways, review logs for usernames containing embedded NUL bytes as a detection signal.

Sources: NVD access.redhat.com ATLAS

What is the risk?

High severity (CVSS 7.1, AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:L/A:N). Exploitation requires network access and possession of a valid, low-privileged RSA-PSK identity, but no user interaction and low attack complexity — an attacker who already holds any PSK credential on the service can attempt the truncation trick to impersonate a different identity. Confidentiality impact is high (unauthorized access to data/sessions belonging to the impersonated identity); integrity impact is low; availability is unaffected. No evidence of active exploitation, no CISA KEV entry, no EPSS score, and no public exploit or scanner template exist yet, which lowers near-term likelihood but does not eliminate risk given the library's broad reach (129 dependents) and history of frequent vulnerabilities (78 prior CVEs in the same package).

How does the attack unfold?

Initial Access

Attacker connects to a gnutls-based RSA-PSK TLS service (e.g., an AI inference gateway) while holding a valid, low-privileged PSK identity.

AML.T0049

Identity Confusion

Attacker sends a crafted PSK username containing a NUL byte that gnutls truncates and wrongly matches against a different, targeted identity.

AML.T0012

Unauthorized Access

Server authenticates the attacker as the impersonated identity, granting a session without possession of the real pre-shared key.

AML.T0040

Impact

Attacker accesses data or services scoped to the impersonated identity, breaching the confidentiality boundary between AI inference components (CVSS C:H).

Initial Access

Attacker connects to a gnutls-based RSA-PSK TLS service (e.g., an AI inference gateway) while holding a valid, low-privileged PSK identity.

AML.T0049

Identity Confusion

Attacker sends a crafted PSK username containing a NUL byte that gnutls truncates and wrongly matches against a different, targeted identity.

AML.T0012

Unauthorized Access

Server authenticates the attacker as the impersonated identity, granting a session without possession of the real pre-shared key.

AML.T0040

Impact

Attacker accesses data or services scoped to the impersonated identity, breaching the confidentiality boundary between AI inference components (CVSS C:H).

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
vLLM	pip	—	No patch
89.7K 128 dependents Pushed 3d ago 24% patched ~54d to patch Full package profile →
discovery/discovery-server-rhel9	—	—	No patch
discovery/discovery-ui-rhel9	—	—	No patch
gnutls	—	—	No patch
gnutls-main	—	—	No patch
libtasn1	—	—	No patch
rhaiis/model-opt-cuda-rhel9	—	—	No patch
rhcos	—	—	No patch
rhui5/cds-rhel9	—	—	No patch
rhui5/haproxy-rhel9	—	—	No patch
rhui5/installer-rhel9	—	—	No patch
rhui5/rhua-rhel9	—	—	No patch

How severe is it?

CVSS 3.1

7.1 / 10

EPSS

1.1%

chance of exploitation in 30 days

Higher than 62% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Moderate

What is the attack surface?

AV Network

AC Low

PR Low

UI None

S Unchanged

C High

I Low

A None

What should I do?

1 step

1) Patch gnutls and libtasn1 to the versions fixed in the referenced Red Hat advisories (RHSA-2026:13274, 20611-20613, 26319, 26409, 29197, 30004, 30849, 30850) across all affected RHEL packages. 2) Rebuild and redeploy any custom images layered on rhaiis/model-opt-cuda-rhel9, rhaiis/vllm-rocm-rhel9, or the discovery-server/discovery-ui images once patched base images are available. 3) Where RSA-PSK authentication is used between AI inference components, consider migrating to certificate-based mTLS if PSK is not strictly required. 4) Audit PSK identity provisioning to ensure no identity's truncated form collides with another identity's full username. 5) Monitor TLS authentication logs on inference gateways and discovery services for PSK usernames containing embedded NUL bytes as an indicator of exploitation attempts.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Auth Bypass Inference AML.T0012 - Valid Accounts AML.T0040 - AI Model Inference API Access AML.T0049 - Exploit Public-Facing Application

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

NIST AI RMF

MANAGE 4.1 - Risk treatment for third-party components

OWASP LLM Top 10

LLM03 - Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2026-42010?

A flaw in gnutls causes RSA-PSK TLS servers to wrongly match a username containing an embedded NUL character against a different, truncated identity, letting an attacker with a valid low-privilege PSK credential authenticate as someone else and bypass intended access controls. This matters because gnutls sits underneath 129 tracked downstream dependents and has racked up 78 other CVEs, and Red Hat's own advisories confirm the bug reaches AI inference infrastructure directly — the RHEL AI Inference Server images (model-opt-cuda-rhel9, vllm-rocm-rhel9) and the discovery-server/discovery-ui components ship the affected library. There is no EPSS score, no CISA KEV listing, no public exploit code, and no Nuclei template as of this writing, so this looks like a quiet library-level bug rather than an actively weaponized one — but CVSS 7.1 (network, low complexity, no user interaction, high confidentiality impact) means it's cheap to exploit once someone builds a proof of concept. Patch gnutls/libtasn1 to the versions fixed in the Red Hat errata (RHSA-2026:13274 and related), rebuild any RHAIIS/vllm-rocm-rhel9 container images from the patched base, and where RSA-PSK is used to authenticate between AI inference nodes or gateways, review logs for usernames containing embedded NUL bytes as a detection signal.

Is CVE-2026-42010 actively exploited?

No confirmed active exploitation of CVE-2026-42010 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-42010?

1) Patch gnutls and libtasn1 to the versions fixed in the referenced Red Hat advisories (RHSA-2026:13274, 20611-20613, 26319, 26409, 29197, 30004, 30849, 30850) across all affected RHEL packages. 2) Rebuild and redeploy any custom images layered on rhaiis/model-opt-cuda-rhel9, rhaiis/vllm-rocm-rhel9, or the discovery-server/discovery-ui images once patched base images are available. 3) Where RSA-PSK authentication is used between AI inference components, consider migrating to certificate-based mTLS if PSK is not strictly required. 4) Audit PSK identity provisioning to ensure no identity's truncated form collides with another identity's full username. 5) Monitor TLS authentication logs on inference gateways and discovery services for PSK usernames containing embedded NUL bytes as an indicator of exploitation attempts.

What systems are affected by CVE-2026-42010?

This vulnerability affects the following AI/ML architecture patterns: model serving, inference infrastructure, service-to-service authentication.

What is the CVSS score for CVE-2026-42010?

CVE-2026-42010 has a CVSS v3.1 base score of 7.1 (HIGH). The EPSS exploitation probability is 1.05%.

What is the AI security impact?

Affected AI Architectures

model servinginference infrastructureservice-to-service authentication

MITRE ATLAS Techniques

AML.T0012 Valid Accounts

AML.T0040 AI Model Inference API Access

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 15

NIST AI RMF: MANAGE 4.1

OWASP LLM Top 10: LLM03

What are the technical details?

Original Advisory

A flaw was found in gnutls. Servers configured with RSA-PSK (Rivest–Shamir–Adleman – Pre-Shared Key) wrongfully matched usernames containing a NUL character with truncated usernames. A remote attacker could exploit this by sending a specially crafted username, leading to an authentication bypass. This vulnerability allows an attacker to gain unauthorized access by circumventing the authentication process.

Exploitation Scenario

An attacker who has obtained (or been issued) a low-privileged RSA-PSK identity on an AI inference gateway crafts a PSK username containing an embedded NUL character designed to be truncated by gnutls into the exact bytes of a different, more privileged identity's username. The gnutls server incorrectly matches the crafted, truncated username against that other identity's record and authenticates the attacker as if they held the real identity's pre-shared key. With that session, the attacker gains unauthorized access to whatever the impersonated identity can reach — potentially internal model inference APIs, the discovery-server/discovery-ui management surface, or data flows intended to be isolated between inference nodes — without ever knowing the actual PSK secret.

Weaknesses (CWE)

CWE-170 Improper Null Termination

CWE-170 — Improper Null Termination: The product does not terminate or incorrectly terminates a string or array with a null character or equivalent terminator.

[Requirements] Use a language that is not susceptible to these issues. However, be careful of null byte interaction errors (CWE-626) with lower-level constructs that may be written in a language that is susceptible.
[Implementation] Ensure that all string functions used are understood fully as to how they append null characters. Also, be wary of off-by-one errors when appending nulls to the end of strings.

Source: MITRE CWE corpus.