Amplifying Latency and Cost of LLM Inference Via Rowhammer Attack

Large language models LLMs achieve high inference efficiency using optimized GPU kernels generated by just-in-time JIT compilation frameworks such as Triton. Triton compiles GPU kernels through an MLIR/LLVM-based pipeline. We investigate whether a Rowhammer-induced bit-flip in the CPU's DRAM can...

Anytime-Valid Detection of LLM Weight Exfiltration

A compromised LLM inference server can leak model weights by encoding payload bits in otherwise plausible token choices. A replay of the same prompt in a trusted server can expose such deviations, but benign numerical nondeterminism also causes token mismatches. Patient attackers can therefore hi...

FragToken: Amplifying LLM Inference Costs through Noncanonical Token Generation

As large language model LLM inference becomes increasingly expensive, resource-consumption attacks pose a growing threat to model providers. Existing attacks typically amplify cost by inducing abnormally long or repetitive outputs on attacker-controlled or triggered requests, making them easier t...

SpliTEE: Improving LLM Inference on Trusted Hardware with Differentially Private GPU Outsourcing

User prompts provided to large language models LLMs may contain sensitive or private information that can be misused by remotely deployed models, such as through inadvertent memorization during retraining. One way to protect user prompts is to execute the LLM inside a trusted execution environmen...

Benchmarking Confidential Computing Performance on NVIDIA Blackwell GPUs

This paper measures the performance impact of running large language model inference and training inside a Trusted Execution Environment TEE on NVIDIA B200 GPUs, using Intel Trust Domain Extensions TDX confidential VMs together with NVIDIA Confidential Computing CC on Blackwell GPUs. The...

EUVD-2026-41925

vLLM denial of service via prompt embeds on M-RoPE models...

PYSEC-2025-58

vLLM is a library for LLM inference and serving. vllm/modelexecutor/weightutils.py implements hfmodelweightsiterator to load the model checkpoint, which is downloaded from huggingface. It uses the torch.load function and the weightsonly parameter defaults to False. When torch.load loads malicious...

CVE-2024-42478

llama.cpp provides LLM inference in C/C++. The unsafe data pointer member in the rpctensor structure can cause arbitrary address reading. This vulnerability is fixed in b3561...

CVE-2024-42479 llama.cpp allows write-what-where in rpc_server::set_tensor

llama.cpp provides LLM inference in C/C++. The unsafe data pointer member in the rpctensor structure can cause arbitrary address writing. This vulnerability is fixed in b3561...

CVE-2024-42478

CVE-2024-42478 concerns llama.cpp where an unsafe data pointer in the rpc_tensor structure can lead to arbitrary address reads. Several sources concur this is a code-level issue; the public CVE description states the vulnerability is fixed in b3561. Some connected advisories also advise upgrading...

CVE-2024-42478 llama.cpp allows Arbitrary Address Read in rpc_server::get_tensor

llama.cpp provides LLM inference in C/C++. The unsafe data pointer member in the rpctensor structure can cause arbitrary address reading. This vulnerability is fixed in b3561...

CVE-2024-42478

llama.cpp provides LLM inference in C/C++. The unsafe data pointer member in the rpctensor structure can cause arbitrary address reading. This vulnerability is fixed in b3561...

CVE-2024-42477

CVE-2024-42477 affects llama.cpp (C/C++) where an unsafe type member in the rpc_tensor structure can trigger a global-buffer-overflow, potentially leaking memory data. The issue is reported as fixed in commit b3561 and is reflected in Fedora security advisory FEDORA-2024-89c69bb9d3 and related ad...

CVE-2024-41130 llama.cpp null pointer dereference in gguf_init_from_file

llama.cpp provides LLM inference in C/C++. Prior to b3427, llama.cpp contains a null pointer dereference in ggufinitfromfile. This vulnerability is fixed in b3427...