Loading vulnerability…
CVE-2025-23320 affects NVIDIA Triton Inference Server on Windows and Linux in the Python backend. According to the provided content, an attacker can send an oversized or very large request that causes the shared memory limit to be exceeded. The resulting verbose error handling discloses the name of an internal IPC/shared memory region used by the Python backend, with examples matching the pattern triton_python_backend_shm_region_*. This is an information disclosure issue in which internal implementation details are exposed through error messages, and the leaked shared memory region identifier can be used as a primitive in a broader exploit chain against Triton deployments prior to version 25.07.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
No public exploits tracked yet. Mallory keeps watching.
No public exploit code observed for this vulnerability.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
8 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
An information disclosure vulnerability in NVIDIA Triton Inference Server that reveals internal shared memory region names through verbose error messages and can be chained with other flaws to facilitate compromise.
A critical vulnerability in NVIDIA Triton Inference Server mentioned as part of a group of 2025 flaws, but not described in detail in the content.
A vulnerability in NVIDIA Triton Inference Server that leaks an internal shared memory region name through verbose error handling when sent an oversized request, serving as the first step in an exploit chain with CVE-2025-23319.
A vulnerability in NVIDIA Triton Inference Server shared memory management related to excessive resource consumption.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.