PT-2026-58762

Name of the Vulnerable Software and Affected Versions NVIDIA Triton Inference Server for Linux affected versions not specified Description An issue exists where an attacker can trigger an uncaught exception. This flaw is network exploitable and requires no authentication, which may lead to a deni...

PT-2026-58761

Name of the Vulnerable Software and Affected Versions NVIDIA Triton Inference Server for Linux affected versions not specified Description An issue exists that allows an attacker to cause uncontrolled resource consumption. This flaw is network exploitable and requires no authentication, which may...

CVE-2026-47478: Use of Expired File Descriptor

NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker can cause the use of an expired file descriptor. A successful exploit of this vulnerability might lead to denial of service...

Evaluating Frontier AI Agents As Autonomous Clinical Security Auditors

Clinical AI models can expose patients to harm when adversarial vulnerabilities go undetected, yet formal security auditing requires statistical expertise, specialized tools, and significant time. We present an open evaluation task, built on METR Task Standard v0.3.0, that tests whether frontier ...

Security Bulletin: NVIDIA Triton Inference Server - July 2026

NVIDIA has released a software update for NVIDIA® Triton Inference Server. To protect your system, clone or update this software to Triton Server r26.05 or later from the NVIDIA Triton Inference Server GitHub repo. Go to NVIDIA Product Security. Details The following table summarizes the potentia...

vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels

SummaryAll temperature validation gates use comparison operators , which silently evaluate to False for NaN and for positive Infinity in Python's IEEE 754 float semantics. Both values pass every guard and propagate to GPU sampling kernels, where they produce undefined behavior or CUDA errors that...

SGLang: Reachable Assertion via  lora_path  in LoRAManager enables remote Denial of Dervice

A security vulnerability has been detected in SGLang 0.5.10.post1. Impacted is an unknown function of the file python/sglang/srt/lora/loramanager.py of the component Inference HTTP Endpoint. Such manipulation of the argument lorapath leads to reachable assertion. The attack can be launched...

PYSEC-2026-3064 SGLang: Reachable Assertion via  lora_path  in LoRAManager enables remote Denial of Dervice

A security vulnerability has been detected in SGLang 0.5.10.post1. Impacted is an unknown function of the file python/sglang/srt/lora/loramanager.py of the component Inference HTTP Endpoint. Such manipulation of the argument lorapath leads to reachable assertion. The attack can be launched...

PYSEC-2026-3060 Amazon SageMaker Python SDK is missing integrity verification in its Triton inference handler

Summary Amazon SageMaker Python SDK is an open-source library for training and deploying machine learning models on Amazon SageMaker. An issue exists where, under certain circumstances, the Triton inference handler deserializes model artifacts without performing integrity verification, allowing...

Securing LLMs in the Wild: Privacy and Security Challenges at the Edge

Large Language Models LLMs are rapidly moving from research settings into the wild, deployed on enterprise infrastructure, personal devices, and edge platforms. While cloud deployments offer scalable compute, concerns over data sovereignty, compliance, latency, and third-party dependence are...

PT-2026-59684

A security vulnerability has been detected in SGLang 0.5.10.post1. Impacted is an unknown function of the file python/sglang/srt/lora/lora manager.py of the component Inference HTTP Endpoint. Such manipulation of the argument lora path leads to reachable assertion. The attack can be launched...

SoK: Federated Learning for Intrusion Detection in Vehicular Networks

Modern vehicular networks face an expanding attack surface across internal Electronic Control Units ECUs and external Vehicle-to-Everything V2X communication. Federated Learning FL has emerged as a decentralized paradigm to deploy Intrusion Detection Systems IDS without compromising data privacy...

PYSEC-2026-3211 Type confusion leading to segfault in Tensorflow

Impact The implementation of shape inference for ConcatV2 can be used to trigger a denial of service attack via a segfault caused by a type confusion: python import tensorflow as tf @tf.function def test: y = tf.rawops.ConcatV2 values=1,2,3,4,5,6, axis = 0xb500005b return y test The axis argument...

Type confusion leading to segfault in Tensorflow

Impact The implementation of shape inference for ConcatV2 can be used to trigger a denial of service attack via a segfault caused by a type confusion:pythonimport tensorflow as [email protected] test: y = tf.rawops.ConcatV2 values=1,2,3,4,5,6, axis = 0xb500005b return ytestThe axis argument is...

PYSEC-2026-3241 Crash when type cannot be specialized in Tensorflow

Impact Under certain scenarios, TensorFlow can fail to specialize a type during shape inference: cc void InferenceContext::PreInputInit const OpDef& opdef, const std::vector& inputtensors, const std::vector& inputtensorsasshapes const auto ret = fulltype::SpecializeTypeattrs, opdef;...

PYSEC-2026-3249 Out of bounds read in Tensorflow

Impact TensorFlow's type inference can cause a heap OOB read as the bounds checking is done in a DCHECK which is a no-op during production: cc if nodet.typeid != TFTUNSET int ix = inputidxi; DCHECKix nodet.argssize "input " i " should have an output " ix " but instead only has " nodet.argssize "...

Out of bounds read in Tensorflow

ImpactTensorFlow's type inference can cause a heap OOB read as the bounds checking is done in a DCHECK which is a no-op during production:ccif nodet.typeid != TFTUNSET int ix = inputidxi; DCHECKix nodet.argssize "input " i " should have an output " ix " but instead only has " nodet.argssize "...

Crash when type cannot be specialized in Tensorflow

ImpactUnder certain scenarios, TensorFlow can fail to specialize a type during shape inference:ccvoid InferenceContext::PreInputInit const OpDef& opdef, const std::vector& inputtensors, const std::vector& inputtensorsasshapes const auto ret = fulltype::SpecializeTypeattrs, opdef;...

Out of bounds read in Tensorflow

Impact The implementation of shape inference for ReverseSequence does not fully validate the value of batchdim and can result in a heap OOB read:pythonimport tensorflow as [email protected] test: y = tf.rawops.ReverseSequence input = 'aaa','bbb', seqlengths = 1,1,1, seqdim = -10, batchdim = -10...

PYSEC-2026-3121 Out of bounds read in Tensorflow

Impact The implementation of shape inference for ReverseSequence does not fully validate the value of batchdim and can result in a heap OOB read: python import tensorflow as tf @tf.function def test: y = tf.rawops.ReverseSequence input = 'aaa','bbb', seqlengths = 1,1,1, seqdim = -10, batchdim = -...