Posts
High-signal AI/security/automation notes.
Filter: All AI Vulnerabilities & CVEs (348)Tools & Comparisons (2)AI Security Research (209)Agent & Supply Chain Security (707)
PyTorch Lightning — PyPI Package Compromised in Mini Shai-Hulud Supply Chain Attack
The popular PyTorch Lightning deep learning framework (31k+ GitHub stars) was compromised with credential-stealing malware in versions 2.6.2 and 2.6.3, part of the ongoing Mini Shai-Hulud campaign targeting AI infrastructure.
VentureBeat — Six Exploits Against AI Coding Agents, All Targeting Credentials
Six independent research teams demonstrated exploits against Codex, Claude Code, Copilot and Vertex AI — every single attack went for the credential, not the model.
Cequence Security — Agent Personas for Scoped MCP Privileges
Cequence Security released Agent Personas in its AI Gateway, solving the agent privilege escalation problem by mapping plain-English role descriptions to scoped virtual MCP endpoints.
DSN 2026 — First Cross-Entity Security Study of the MCP Ecosystem
University of Delaware researchers scanned 67,000+ MCP servers across six registries, revealing widespread server hijacking risks, tool metadata manipulation, and missing verification in MCP hosts.
MCPTox — Tool Poisoning Benchmark Shows 73% Attack Success Rate on MCP Agents
MCPTox benchmark evaluates 20 LLM agents against tool poisoning attacks on 45 real MCP servers — o1-mini hit 72.8% success rate, with more capable models often more vulnerable.
TeamPCP Escalates Mini Shai-Hulud Campaign to SAP npm and PyTorch Lightning
The TeamPCP supply-chain operation expanded from SAP npm packages to PyPI (PyTorch Lightning) and intercom-client within 48 hours, targeting AI and ML developer ecosystems.
Novee — Cursor IDE CVE-2026-26268: Git Hooks Enable RCE via AI Coding Agent
Novee researchers discovered a high-severity vulnerability in Cursor IDE (CVE-2026-26268, CVSS 8.1) where the AI coding agent autonomously triggers malicious Git hooks when opening attacker-controlled repositories.
NSFOCUS — AI-Scan Security Scanner for OpenClaw Ecosystem
NSFOCUS announced AI-Scan, a dedicated security scanner for the OpenClaw agent ecosystem covering CVE matching, credential leak detection, memory poisoning, and supply-chain risk.
RAGShield — Numerical Claim Manipulation in RAG Systems Evades Embedding Defenses
New research proves embedding-based RAG defenses miss 79-90% of numerical manipulations — changing a tax figure by $50,000 yields 0.9998 cosine similarity, invisible to detection.
SecureAuth — Agent Trust Registry Opens to the Public for AI Agent Governance
Wiz — Red Agent, AI-BOM, and Wiz Code Expand AI Application Security Platform
Wiz launches Red Agent for AI-driven vulnerability simulation, AI-BOM for framework inventory, and Wiz Code guardrails — expanding its AI Application Protection Platform from code to cloud edge.
ARMO — Why AI Supply Chain Scanning Misses Half the Risk
Most AI security scanners only cover packaged dependencies, missing the dynamic components agents load at runtime — model adapters, MCP connections, and first-invocation tools.
arXiv — LLMs Fail at Open-Ended Threat Hunting (3.8% Best Score)
A new benchmark tests five frontier LLMs on real-world threat hunting across 106 attack procedures — the best model flags only 3.8% of malicious events.
Lufsec — Model Context Protocol: 4 Trust Boundaries Attackers Are Exploiting
Four MCP trust boundaries — permissions, tools, inputs, and observability — that teams are getting wrong when connecting AI agents to real systems.
DPRK PromptMink — Claude Opus Used to Insert Malicious npm Dependency
Hugging Face LeRobot — Critical Pickle Deserialization RCE (CVE-2026-25874)
Pipecat Voice Agent Framework — Pickle Deserialization RCE (CVE-2025-62373)
Critical RCE in Pipecat, the open-source Python framework for voice and multimodal AI agents, via insecure pickle deserialization in LivekitFrameSerializer.
T-MAP — Red-Teaming LLM Agents with Trajectory-aware Evolutionary Search
New arXiv paper introduces T-MAP, a red-teaming framework that achieves 57.8% attack success rate against frontier LLM agents through trajectory-aware evolutionary search in MCP environments.