EUVD-2026-88961
GHSA-wrcm-gqj2-2cfj
Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessObtain local low-priv access to DeepStream host
-
DeliveryCraft malicious tensor dimensions in YAML config
-
ExploitPipeline parses config triggering integer overflow
-
ExecutionUndersized allocation / bounds miscalculation
-
ImpactCrash, memory disclosure, or data tampering
Vulnerability Assessment
| Weakness |
CWE
ASVS 5.0
V15 - Secure Coding and Architecture Integer overflow checks
MITRE ATT&CK
1 technique across 1 tactic: Execution The technique list with ATT&CK links and the kill-chain view is part of the Basic plan and up. |
| Exploitation Conditions | Requires the attacker to control the tensor-dimension values inside a YAML configuration file that the DeepStream inference pipeline loads - i.e. local access with the ability to supply or modify pipeline config (consistent with AV:L/PR:L). Exploitation is LIMITED by: no network vector (attacker must already be local/authenticated on the DeepStream host), and dependence on the deployment feeding attacker-influenced YAML into the pipeline. Environments that only load trusted, operator-authored configs are not exposed; the risk rises where config content is sourced from untrusted users or remote/automated provisioning. |
| Impact | The provided CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H, base 7.8) describes a LOCAL, low-privilege, no-interaction flaw with full C/I/A impact - a real severity ceiling but gated hard by the AV:L / PR:L requirement that the attacker already have local, authenticated access AND the ability to supply or modify the YAML configuration file the DeepStream pipeline consumes. This is not a remotely-reachable network vulnerability, so it is not a mass-exploitation or worm candidate. No EPSS score, no CISA KEV listing, and no public exploit code were provided, so exploitation status is: no public exploit identified at time of analysis. The 7.8 'High' rating is driven entirely by the triple-High impact; in practical terms this is a lower real-world priority than the number suggests because the trigger is a config-supply/local-access scenario rather than an internet-facing entry point. Note a mild tension worth verifying: the description frames the bug as 'supplying crafted tensor dimensions in a YAML configuration file,' which fits AV:L (attacker controls a local file), but any deployment that ingests DeepStream config from an untrusted or remote source could raise the effective attack surface - verify how configs are provisioned in the specific environment. |
| Exploit ScenarioDerived | An attacker with local, low-privilege access to a host running NVIDIA DeepStream (or the ability to plant/modify the pipeline's YAML config) crafts tensor dimension values that trigger an integer overflow (CWE-190) during config parsing/tensor allocation. The resulting undersized allocation or miscalculated bounds leads to denial of service, disclosure of adjacent memory, or data tampering in the inference pipeline. No public proof-of-concept was provided in the intelligence sources. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Sep 29, 2026
Integer overflow in NVIDIA DeepStream is reachable when crafted tensor dimensions are supplied in a YAML configuration file loaded by the inference pipeline, enabling an authenticated local attacker to cause denial of service, information disclosure, and data tampering. Exploitation requires local access and the ability to modify or supply the DeepStream pipeline YAML, consistent with the CVSS 3.1 vector AV:L/AC:L/PR:L/UI:N, so deployments that only load trusted operator-authored configurations are not exposed. No public exploit code or confirmed active exploitation (CISA KEV) was identified in the provided intelligence at time of analysis.
RemediationAI
Check NVIDIA's advisory at https://github.com/NVIDIA/product-security/tree/main/2026/5886 for fixed-release guidance, because no exact patched version was identified in the provided data; apply that vendor-released update as soon as it is confirmed. Until a fix is applied, restrict write access to DeepStream pipeline YAML files and their directories so only trusted operators can provide tensor dimensions, validate or cap dimension values before the pipeline loads them, and run DeepStream under a dedicated low-privilege account. Environments that ingest YAML from untrusted users or automated provisioning should treat those inputs as untrusted, reject malformed or oversized dimension values, and isolate the DeepStream process from sensitive data; these controls can disrupt legitimate dynamic-configuration and automated-provisioning workflows, and least-privilege execution may affect access to GPUs or device nodes.
Affected VersionsENISA
Affects your stack
Find out whether the stacks and servers you run are on an affected version.
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
NVIDIA DeepStream is a streaming analytics toolkit used to build AI video inference pipelines, commonly with GStreamer and TensorRT components. The vulnerable path involves parsing YAML configuration files that define model and tensor dimensions; if integer values from that file are used in size or offset arithmetic without safe validation, a CWE-190 integer overflow can occur, leading to incorrect buffer sizing or memory corruption and high confidentiality, integrity, and availability impact. The supplied CPE cpe:2.3:a:nvidia:deepstream:*:*:*:*:*:*:*:* uses a wildcard version, so exact affected and fixed releases are not enumerated in the input data. The CVSS vector contains AV:L and PR:L, confirming the attacker must already have local, authenticated access and must be able to control the YAML loaded by the inference pipeline; no network-only vector is described.
Know when Deepstream gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H
Attacker supplies a local YAML config (AV:L, PR:L, no UI); integer overflow yields memory disclosure, tampering, and crashes, supporting C:H/I:H/A:H with no scope change.
Primary rating from Vendor (nvidia).
CVSS VectorVendor: nvidia
DescriptionCVE.org
NVIDIA DeepStream contains a vulnerability where an attacker could cause an integer overflow by supplying crafted tensor dimensions in a YAML configuration file. A successful exploit of this vulnerability might lead to denial of service, information disclosure, data tampering.
Executive ViewAI
NVIDIA DeepStream, the SDK used to build the AI video-analytics and inference pipelines that power many of our vision-based workloads, contains an unpatched integer-overflow flaw (CVE-2026-65102) that can crash inference services, expose data processed in those pipelines, and allow tampering with results. Exploitation is not remote or anonymous: it requires an attacker who already has a local foothold on the host and the ability to modify or supply the pipeline configuration file, so exposure is concentrated in environments where operators, contractors, or third-party integrators have write access to DeepStream YAML configurations. Because no vendor fix exists yet, leadership should treat this as a contained but unresolved risk requiring configuration hardening and access controls rather than emergency outage-driven patching.
No vendor-released patch identified at time of analysis. Interim mitigations: restrict write access to DeepStream pipeline YAML configuration files and directories to a minimal set of trusted, authenticated operators; run DeepStream with least privilege as a non-root service account; mount pipeline configurations read-only and pin container images by digest; enable file-integrity monitoring or hashing on configuration paths to detect unauthorized changes; validate or bound tensor dimensions before pipeline load; and limit local shell access on inference hosts. No public exploit identified at time of analysis.
Threat Intelligence Overview
NVD
CVE.org
EPSS
Vulnrichment
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Integer overflow occurs when an arithmetic operation produces a result that exceeds the maximum value a given integer type can store.
Information disclosure occurs when an application unintentionally exposes sensitive data that aids attackers in reconnaissance or directly compromises security.
Denial of Service attacks render applications or systems unavailable by overwhelming resources or triggering failure conditions.
Exploit-DB