Disclosures

The attack techniques, behaviors, and vulnerabilities uncovered through these disclosures feed Mindgard’s proprietary knowledge base and continuously strengthen our AI Security Platform.

156
Vulnerabilities reported
Submitted to affected vendors
132
Vulnerabilities accepted
Confirmed valid by vendors
98
Vulnerabilities patched
Vendor fix confirmed
24
Affected vendors
Distinct organizations
38
Affected products
Distinct AI systems & tools

Disclosures reported over time

Cumulative count by report date · mid-2025 to today
Cumulative

Affected vendors

Block area = number of disclosure records
156 records

All Disclosures

Published findings and vulnerabilities currently under coordinated disclosure.

Upcoming
Published
Title
Vendor
Status
Published
NVIDIA TensorRT-LLM NIXL RDMA Write to Attacker-Controlled GPU Addresses via Notification MemoryDesc
NVIDIA
Upcoming
NVIDIA TensorRT-LLM Unauthenticated Disagg Cluster Storage Enables Fake Worker Registration -> Request Hijacking
NVIDIA
Upcoming
NVIDIA TensorRT-LLM Triton Backend context_phase_params Deserialization Enables SSRF + RDMA Injection
NVIDIA
Upcoming
NVIDIA TensorRT-LLM Missing HMAC Key in trtllm-serve Disaggregated Launch Path
NVIDIA
Upcoming
NVIDIA TensorRT Arbitrary Code Execution via Embedded Plugin dlopen
NVIDIA
Upcoming
NVIDIA TensorRT Unbounded nbDims in Tensor Shape Deserialization Stack Buffer Overflow
NVIDIA
Upcoming
NVIDIA TensorRT Unchecked Shader Index in readEngineFromArchive Heap OOB Read
NVIDIA
Upcoming
NVIDIA TensorRT-LLM Unauthenticated Disaggregated opaque_state Deserialization SSRF + KV Cache Data Injection
NVIDIA
Upcoming
Joblib Model File Scanner Bypass via Compressed Files
Joblib
Upcoming
Joblib Model File Path Traversal via NDArrayWrapper / ZNDArrayWrapper in Compat Format
Joblib
Upcoming
Joblib Model File Pickle Restriction Bypass via Hidden Payload in Array Data
Joblib
Upcoming
Microsoft ONNX Runtime Code Execution via Tile Operator Heap Buffer Overflow
Microsoft
Upcoming
Anyscale Ray Serve Internal HTTP Header Forging
Anyscale
Upcoming
Anyscale Ray Serve Cross-Site Scripting via X-Request-Id Response Header Reflection
Anyscale
Upcoming
Anyscale Ray Serve Code Execution via Replica gRPC Pickle Deserialization
Anyscale
Upcoming
Title
Vendor
Status
Published
Microsoft Azure Prompt Shield Guardrail Evasion
Microsoft
Published
June 24, 2024

See What Your AI Defenses Miss

See how Mindgard exposes and fixes exploitable AI risk across your AI agents and systems.

Book a Demo

Mindgard helps organizations discover, assess and defend their AI systems. Spun out of more than a decade of AI security research at Lancaster University in the UK and headquartered in Boston and London, Mindgard operationalizes the expertise of AI researchers and offensive security practitioners through a Security Platform that performs Shadow AI discovery, AI red teaming, and run-time AI protection to assess and mitigate risk across models, agents, and applications.

AICPA SOC SOC 2 Type 2 Compliant