NVIDIA has patched two high‑severity flaws in its Triton Inference Server for Linux that could let remote attackers crash AI inference services using specially crafted malicious payloads.
The issues affect all Triton Inference Server releases on Linux before version r25.10. NVIDIA urges customers to upgrade immediately and review its Secure Deployment Considerations Guide to reduce exposure in production environments.
Two High-Severity CVEs Leading to Denial of Service
The update fixes CVE-2025-33211 and CVE-2025-33201, both rated 7.5 (High) under the CVSS v3.1 scoring system.
In CVE-2025-33211, Triton for Linux fails to properly validate a quantity field in user input. In simple terms, the server trusts some numeric value from the client without enough checks.
An attacker can send a crafted request that abuses this weak validation and causes the server process to stop responding, resulting in a denial-of-service (DoS) condition.
This issue is categorized under CWE-1284 (Improper Validation of Specified Quantity in Input).
CVE-2025-33201 involves an improper check for unusual or exceptional conditions when handling extra large payloads.
By sending abnormally large or malformed request data to the Triton Inference Server, an attacker can trigger an unrecoverable error or resource exhaustion, again leading to a DoS state.
This vulnerability falls under CWE-754, which covers failures to properly handle exceptional conditions.
For both flaws, the CVSS vector AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H means they are exploitable over the network, require low attack complexity, need no privileges and no user interaction, and primarily impact availability (high impact) while leaving confidentiality and integrity unaffected.
Both vulnerabilities affect Triton Inference Server on Linux, all versions prior to r25.10.
NVIDIA has addressed the issues in version r25.10, now available on the Triton Inference Server Releases page on GitHub.
Organizations running Triton in production, especially in internet-exposed or multi-tenant environments, should:
- Upgrade Triton Inference Server to r25.10 or later.
- Review NVIDIA’s Secure Deployment Considerations Guide for hardened configurations.
- Monitor logs and observability tools for unusual spikes in request size or sudden Triton process crashes that might indicate exploitation attempts.
NVIDIA credits seaw1nd for reporting CVE-2025-33211, and Tyler Zars and Rob Blakely (Technical Debt Collectors – Trend Micro ZDI) along with taiphung217 (MDPC, OneMount Group) for CVE-2025-33201.
Find this Story Interesting! Follow us on Google News, LinkedIn and X to Get More Instant Updates
%20(1)%20(1).webp?fit=1600,900&ssl=1&description=The issues affect all Triton Inference Server releases on Linux before version r25.10. NVIDIA urges customers to upgrade immediately){kind=link}