Important: Red Hat Security Advisory: Red Hat AI Inference Server 3.2.2 (rocm)

Red Hat AI Inference Server 3.2.2 rocm is now available. Red Hat® AI Inference Server...

Important: Red Hat Security Advisory: Red Hat AI Inference Server 3.2.2 (cuda)

Red Hat AI Inference Server 3.2.2 cuda is now available. Red Hat® AI Inference Server...

Important: Red Hat Security Advisory: Red Hat AI Inference Server Model Optimization Tools 3.2.2 (CUDA)

Red Hat AI Inference Server Model Optimization Tools 3.2.2 CUDA is now available. Red Hat® AI Inference Server Model Optimization Tools...

Link Inference Attack on Privacy-Preserving Knowledge Graphs

Knowledge Graphs KGs are widely used to store and share structured information across sensitive domains such as healthcare, fi- nance, and social networks. A common privacy practice is to delete sen- sitive relations before publishing the graph, under the assumption that removing edges is...

PT-2026-103286

LightLLM through 1.2.0 mounts reinforcement learning control routes on the public HTTP API without authentication checks. Unauthenticated attackers can call endpoints like /pause generation, /abort request, /flush cache, and /init weights update group to disrupt inference operations and wedge...

CVE-2026-103270: Missing Authentication for Critical Function

LightLLM through 1.2.0 mounts reinforcement learning control routes on the public HTTP API without authentication checks. Unauthenticated attackers can call endpoints like /pausegeneration, /abortrequest, /flushcache, and /initweightsupdategroup to disrupt inference operations and wedge workers o...

argostranslate (>=1.4.0, <=1.9.5), audim (>=0.0.6, <=0.0.7) +64 more potentially affected by CVE-2026-102567 via ctranslate2 (>=1.20.1, <=4.7.2)

ctranslate2 PYPI version =1.20.1, =1.4.0, =0.0.6, =0.2.1, =2.1.0, =0.0.1, =1.4.0, =0.0.9, =0.1.0, =0.2.0, =0.3.1, =0.0.15, =0.0.30 and more Source cves: CVE-2026-102567 Source advisory: MEND:CVE-2026-102567...

Deep Learning Latency Attacks and Defenses: A Cross-Domain Survey of Availability Threats

Adversarial machine learning has focused mainly on integrity, but availability is an increasingly consequential complement. Latency attacks also energy-latency attacks increase inference-time work, energy, or response time, causing deadline misses, throughput collapse, or resource exhaustion in...

Inference-Layer Security: Defending against Adversarial Inference and Infrastructure Abuse

A Technical Report: Operating a large language model LLM as a service requires more than inference infrastructure: the provider must also defend against adversarial interactions that seek to exploit the service, including jailbreaking for harmful use, sophisticated denial of service, and...

GAZEleak: Passcode Inference against Eye-Tracking XR Devices through External Observation

Mixed-reality headsets such as Apple Vision Pro replace the touch screen with gaze-as-pointer interaction: the wearer looks at a target and confirms with an air pinch. Because the display is inside the headset and the eye tracker is walled off from third-party software, such input is widely assum...

CounterSteer: Suppressing Indirect Prompt Injection with Activation Steering

Indirect prompt injection makes an LLM agent treat untrusted retrieved text as instructions. We present CounterSteer, an inference-time defense that suppresses this behavior inside the model. Per model, a five-step recipe fits a residual-stream direction from paired episodes differing only in...

evoke-vllm (>=0.1.0, <=0.1.1), lazyllm-verl (0.3.2.dev1) +6 more potentially affected by CVE-2026-100653 via vllm (>=0.22.1, <=0.27.1)

vllm PYPI version =0.22.1, =0.1.0, =2.8.0, =2.2.2, =0.1.10, =0.1.0, =0.1.3 Source cves: CVE-2026-100653 Source advisory: SNYK:PYTHON-VLLM-20158365...

lazyllm-verl (0.3.2.dev1), londonaicentre-mesa-local (>=2.8.0, <=2.8.1) +1 more potentially affected by CVE-2026-100652 via vllm (>=0.22.0, <=0.23.0)

vllm PYPI version =0.22.0, =2.8.0, =2.1.3, =2.2.4 Source cves: CVE-2026-100652 Source advisory: SNYK:PYTHON-VLLM-20158366...

lazyllm-verl (0.3.2.dev1), londonaicentre-mesa-local (>=2.8.0, <=2.8.1) +1 more potentially affected by CVE-2026-100652 via vllm (>=0.22.0, <=0.23.0)

vllm PYPI version =0.22.0, =2.8.0, =2.1.3, =2.2.4 Source cves: CVE-2026-100652 Source advisory: OSV:PYSEC-2026-4185...

CVE-2026-100651

A flaw was found in vLLM. The service fails to properly enforce length limits on input prompts when handling certain multimodal inference requests. A remote attacker with access to the inference service could exploit this vulnerability by submitting an excessively long prompt, causing an internal...

lazyllm-verl (0.3.2.dev1), londonaicentre-mesa-local (>=2.8.0, <=2.8.1) +1 more potentially affected by CVE-2026-100652 via vllm (>=0.22.0, <=0.23.0)

vllm PYPI version =0.22.0, =2.8.0, =2.1.3, =2.2.4 Source cves: CVE-2026-100652 Source advisory: MEND:CVE-2026-100652...

CVE-2026-100651 vllm before 0.29.0 Denial of Service via Decoder Prompt Length Bypass

vLLM before 0.29.0 fails to enforce decoder prompt-length validation on the disaggregated serving endpoint /inference/v1/generate. When the request contains a 'features' multimodal payload, vllm/entrypoints/serve/disagg/serving.py builds a multimodal EngineInput directly from the caller-supplied...

CVE-2026-100651

vLLM before 0.29.0 fails to enforce decoder prompt-length validation on the disaggregated serving endpoint /inference/v1/generate. When the request contains a 'features' multimodal payload, vllm/entrypoints/serve/disagg/serving.py builds a multimodal EngineInput directly from the caller-supplied...

quail-engine (0.1.0), sendnn-inference (>=2.5.0, <=2.6.3) +2 more potentially affected by CVE-2026-100649 via vllm (>=0.25.1, <=0.28.0)

vllm PYPI version =0.25.1, =2.5.0, =0.1.0, =0.1.4 Source cves: CVE-2026-100649 Source advisory: MEND:CVE-2026-100649...

Leaky Students: Membership Inference against On-Policy Distillation

On-policy distillation OPD trains a student to match a teacher's next-token distributions on student-generated trajectories. However, privileged information supplied to the teacher for OPD training may contain sensitive data. Whether the student leaks private information about the records supplie...