EUVD-2026-98548
GHSA-3rv3-g2r4-2v82
Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessCraft malicious input file or dataset
-
ExploitVictim loads it into DALI pipeline
-
ExecutionTrigger out-of-bounds read in parser
-
ImpactLeak process memory or crash service
Vulnerability Assessment
| Weakness |
CWE
ASVS 5.0
V15 - Secure Coding and Architecture Bounds checking, read limits
MITRE ATT&CK
1 technique across 1 tactic: Collection The technique list with ATT&CK links and the kill-chain view is part of the Basic plan and up. |
| Exploitation Conditions | Exploitation requires an application or workflow that uses NVIDIA DALI to parse attacker-supplied input, and a low-privileged local user (PR:L) must actively process that crafted input (UI:R) through a DALI data-loading/augmentation pipeline. Limiting factors: the attack vector is local (AV:L), not network-reachable; it is not exploitable against an idle default install without a victim loading malicious data; privileges and user interaction are both required, meaning an unauthenticated remote attacker cannot trigger it directly. Realistic exposure is highest where DALI processes untrusted or externally-sourced datasets/media. |
| Impact | The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H, base 7.3 HIGH) describes a locally-exploitable flaw requiring low privileges AND user interaction - not a remotely-reachable issue. No EPSS score, CISA KEV listing, or POC was provided in the input, so there is no public exploit identified at time of analysis and no evidence of active exploitation. The combination of local attack vector plus required user interaction substantially lowers real-world urgency compared to the 7.3 headline: an attacker must get a victim to process attacker-controlled input through a DALI pipeline. Note a tension between the CWE and the impact: CWE-125 is an out-of-bounds READ, which naturally maps to information disclosure (C) and denial of service (A), yet the vendor assigns I:H (data tampering) - a read primitive does not straightforwardly corrupt data, so the integrity rating should be verified against NVIDIA's advisory. Overall this is a moderate-priority, patch-in-normal-cycle issue for teams running DALI on untrusted input, not an emergency. |
| Exploit ScenarioDerived | An attacker crafts a malicious input (e.g., a corrupted image, encoded file, or dataset element) and gets a victim to feed it into an application built on NVIDIA DALI. When DALI parses the crafted data, it performs an uncontrolled out-of-bounds read, leaking adjacent process memory or crashing the pipeline. No public proof-of-concept exploit was identified in the provided data. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Oct 13, 2026
An uncontrolled out-of-bounds read in NVIDIA DALI (Data Loading Library) can be triggered when a low-privileged local user processes specially crafted input through a DALI data-loading or augmentation pipeline (CVSS 3.1 base 7.3, AV:L/AC:L/PR:L/UI:R), resulting in memory disclosure, data tampering, and denial of service. Impact is bounded by the local attack vector and the requirement for both low privileges and user interaction - an unauthenticated remote attacker cannot trigger it directly, and an idle default installation is not exploitable. Exposure is greatest in workflows where DALI ingests untrusted or externally sourced datasets and media; the CPE record (cpe:2.3:a:nvidia:dali:*:*:*:*:*:*:*:*) indicates the issue is not version-scoped in the supplied data, and no public exploit code or CISA KEV entry has been identified at time of analysis.
RemediationAI
Apply the fixed NVIDIA DALI release identified in NVIDIA's product security advisory for CVE-2026-65150 at https://github.com/NVIDIA/product-security/tree/main/2026/5900 - the supplied data points to this advisory as the remediation source but does not name an exact patched version, so obtain the version number from that page before planning an upgrade and pin it in your dependency manifest (requirements/conda/Docker base image). If immediate upgrade is not possible, the most effective compensating control is to stop feeding untrusted or externally sourced datasets and media into DALI pipelines: pre-validate and re-encode inputs with a hardened decoder, or restrict ingestion to curated internal datasets, accepting a loss of flexibility in data sourcing. Run DALI-consuming workloads as an unprivileged, sandboxed process or container with a read-only root filesystem, seccomp/AppArmor confinement, and no access to sensitive files or host secrets, which limits the value of any leaked adjacent memory and contains crash-based denial of service at the cost of some performance and operational complexity. Where the pipeline is multi-tenant, isolate workloads per tenant and cap memory so an out-of-bounds read that crashes the process cannot degrade other jobs. Because exploitation requires a local low-privileged actor plus a victim actively processing crafted input (AV:L, PR:L, UI:R), reducing the number of local accounts able to submit data to DALI workflows and monitoring those pipelines for abnormal memory-error crashes are practical interim measures, but neither removes the underlying CWE-125 defect.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
NVIDIA DALI is a GPU-accelerated data loading and augmentation library used to build input pipelines for deep learning frameworks such as PyTorch and TensorFlow, decoding and transforming images, video, audio, and other dataset formats. The root cause falls under CWE-125 (Out-of-bounds Read): a length, index, or offset derived from attacker-controlled input is used to access memory beyond the bounds of an allocated buffer, so the process reads adjacent heap or stack contents rather than failing cleanly. Because DALI decoders and operators run inside the victim application's process, the read can leak sensitive adjacent memory while also producing corrupt downstream tensors (integrity impact) and raising the risk of a crash-induced denial of service. The CPE string cpe:2.3:a:nvidia:dali:*:*:*:*:*:*:*:* does not constrain a specific version, so the affected range should be taken from the vendor advisory rather than inferred here. The CVSS vector's AV:L component reflects that the crafted input must be delivered to a local DALI pipeline rather than over the network, and PR:L/UI:R confirms that both a low-privileged actor and a victim's active processing step are prerequisites; our independent assessment narrows the integrity impact to Low relative to the NVD-published H, while retaining High confidentiality and availability impact.
Know when Dali 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:R/S:U/C:H/I:H/A:H
Local pipeline input with low-priv user and required interaction gives AV:L/PR:L/UI:R; an OOB read leaks memory (C:H) and crashes processing (A:H), but integrity impact is limited (I:L) since a read does not directly write data.
Primary rating from Vendor (nvidia).
CVSS VectorVendor: nvidia
DescriptionCVE.org
NVIDIA DALI contains a vulnerability where an attacker could cause an uncontrolled out-of-bounds read. A successful exploit of this vulnerability mighty lead to information disclosure, data tampering, and denial of service.
Articles & Coverage 1
Executive ViewAI
NVIDIA DALI, a data-loading and augmentation component used in AI/ML pipelines, exposes a high-severity memory-safety flaw that a low-privileged local user can trigger by processing crafted input, potentially causing memory disclosure, data tampering, or denial of service. The organizational risk is greatest for teams that feed untrusted or externally sourced datasets and media into DALI-based pipelines, where pipeline integrity and availability could be affected. No public exploit identified at time of analysis, but no vendor-released patch is currently identified, so leadership should ensure interim controls are in place while monitoring NVIDIA guidance.
No vendor-released patch identified at time of analysis; interim mitigations include restricting DALI processing to trusted datasets, enforcing least privilege and separation of duties for local users, sandboxing or containerizing DALI pipelines, monitoring for memory-disclosure or integrity anomalies, and applying network/filesystem isolation until an NVIDIA fix is available.
Threat Intelligence Overview
NVD
CVE.org
EPSS
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
A buffer overflow occurs when a program writes more data to a memory buffer than it was allocated to hold, causing the excess data to spill into adjacent memory regions.
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.
An uncontrolled out-of-bounds write in NVIDIA DALI (the Data Loading Library used for GPU-accelerated data preprocessing
An uncontrolled out-of-bounds read in NVIDIA DALI is triggered while the library parses attacker-supplied data inside a
Arbitrary code execution in NVIDIA DALI (all versions prior to 2.0) allows local authenticated attackers with low privil
Heap-based buffer overflow in NVIDIA DALI enables local authenticated attackers to achieve code execution, data tamperin
Local code execution and data tampering in NVIDIA DALI (Data Loading Library) is possible when a low-privileged user is
Stack-based buffer overflow in NVIDIA DALI lets a locally authenticated, low-privileged user who can get a crafted input
Exploit-DB