PYSEC-2025-53

vLLM is an inference and serving engine for large language models LLMs. Prior to version 0.9.0, when a new prompt is processed, if the PageAttention mechanism finds a matching prefix chunk, the prefill process speeds up, which is reflected in the TTFT Time to First Token. These timing differences...

Bayesian Perspective on Memorization and Reconstruction

We introduce a new Bayesian perspective on the concept of data reconstruction, and leverage this viewpoint to propose a new security definition that, in certain settings, provably prevents reconstruction attacks. We use our paradigm to shed new light on one of the most notorious attacks in the...

Confidential Guardian: Cryptographically Prohibiting the Abuse of Model Abstention

Cautious predictions -- where a machine learning model abstains when uncertain -- are crucial for limiting harmful errors in safety-critical applications. In this work, we identify a novel threat: a dishonest institution can exploit these mechanisms to discriminate or unjustly deny services under...

Practical Bayes-Optimal Membership Inference Attacks

We develop practical and theoretically grounded membership inference attacks MIAs against both independent and identically distributed i.i.d. data and graph-structured data. Building on the Bayesian decision-theoretic framework of Sablayrolles et al., we derive the Bayes-optimal membership...

GHSA-VRQ3-R879-7M65 vLLM Tool Schema allows DoS via Malformed pattern and type Fields

Summary The vLLM backend used with the /v1/chat/completions OpenAPI endpoint fails to validate unexpected or malformed input in the "pattern" and "type" fields when the tools functionality is invoked. These inputs are not validated before being compiled or parsed, causing a crash of the inference...

TensorShield: Safeguarding On-Device Inference by Shielding Critical DNN Tensors with TEE

To safeguard user data privacy, on-device inference has emerged as a prominent paradigm on mobile and Internet of Things IoT devices. This paradigm involves deploying a model provided by a third party on local devices to perform inference tasks. However, it exposes the private model to two primar...

Unveiling Impact of Frequency Components on Membership Inference Attacks for Diffusion Models

Diffusion models have achieved tremendous success in image generation, but they also raise significant concerns regarding privacy and copyright issues. Membership Inference Attacks MIAs are designed to ascertain whether specific data were utilized during a model's training phase. As current MIAs...

AdInject: Real-World Black-Box Attacks on Web Agents Via Advertising Delivery

Vision-Language Model VLM based Web Agents represent a significant step towards automating complex tasks by simulating human-like interaction with websites. However, their deployment in uncontrolled web environments introduces significant security vulnerabilities. Existing research on adversarial...

BitHydra: Towards Bit-Flip Inference Cost Attack against Large Language Models

Large language models LLMs have shown impressive capabilities across a wide range of applications, but their ever-increasing size and resource demands make them vulnerable to inference cost attacks, where attackers induce victim LLMs to generate the longest possible output content. In this paper,...

Engineering Trustworthy Machine-Learning Operations with Zero-Knowledge Proofs

As Artificial Intelligence AI systems, particularly those based on machine learning ML, become integral to high-stakes applications, their probabilistic and opaque nature poses significant challenges to traditional verification and validation methods. These challenges are exacerbated in regulated...

Strong Membership Inference Attacks on Massive Datasets and (Moderately) Large Language Models

State-of-the-art membership inference attacks MIAs typically require training many reference models, making it difficult to scale these attacks to large pre-trained language models LLMs. As a result, prior research has either relied on weaker attacks that avoid training reference models e.g.,...

CVE-2024-0095

NVIDIA Triton Inference Server for Linux and Windows contains a vulnerability where a user can inject forged logs and executable commands by injecting arbitrary data as a new log entry. A successful exploit of this vulnerability might lead to code execution, denial of service, escalation of...

CVE-2024-0103

NVIDIA Triton Inference Server for Linux contains a vulnerability where a user may cause an incorrect Initialization of resource by network issue. A successful exploit of this vulnerability may lead to information disclosure...

CVE-2024-0100

NVIDIA Triton Inference Server for Linux contains a vulnerability in the tracing API, where a user can corrupt system files. A successful exploit of this vulnerability might lead to denial of service and data tampering...

CVE-2024-0088

NVIDIA Triton Inference Server for Linux contains a vulnerability in shared memory APIs, where a user can cause an improper memory access issue by a network API. A successful exploit of this vulnerability might lead to denial of service and data tampering...

CVE-2023-31036

NVIDIA Triton Inference Server for Linux and Windows contains a vulnerability where, when it is launched with the non-default command line option --model-control explicit, an attacker may use the model load API to cause a relative path traversal. A successful exploit of this vulnerability may lea...

CVE-2022-21731

Tensorflow is an Open Source Machine Learning Framework. 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. The axis argument is translated into concatdim in the ConcatShapeHelper helper function. Then, a...

CVE-2022-41914

Zulip is an open-source team collaboration tool. For organizations with System for Cross-domain Identity ManagementSCIM account management enabled, Zulip Server 5.0 through 5.6 checked the SCIM bearer token using a comparator that did not run in constant time. Therefore, it might theoretically be...

Towards Anonymous Neural Network Inference

We introduce funion, a system providing end-to-end sender-receiver unlinkability for neural network inference. By leveraging the Pigeonhole storage protocol and BACAP blinding-and-capability scheme from the Echomix anonymity system, funion inherits the provable security guarantees of modern...

An Attack to Break Permutation-Based Private Third-Party Inference Schemes for LLMs

Recent advances in Large Language Models LLMs have led to the widespread adoption of third-party inference services, raising critical privacy concerns. Existing methods of performing private third-party inference, such as Secure Multiparty Computation SMPC, often rely on cryptographic methods...