What changed in AI security, May 4 to May 10, 2026
May 4 to May 10, 2026 (ISO week 2026-W19). Weeks run Monday to Sunday in UTC.
189 records published, +7 on the previous week: 50 vulnerabilities (+19), 0 incidents (no change), 17 research items (-3), 120 news items (-11), 2 policy items (+2).
Critical and high advisories
Vulnerability records rated critical or high, newest first. Showing 25 of 33.- High
CVE-2026-41705: Spring AI MilvusVectorStore filter-expression injection in delete
CVE-2026-41705NVD/CVE Database - High
CVE-2026-44286: FastGPT unauthenticated SSRF through lafModule fetchData in workflow node
CVE-2026-44286NVD/CVE Database - High
CVE-2026-42345: FastGPT is an AI Agent building platform. In versions 4.14.11 and prior, FastGPT's isInternalAddress() function in…
CVE-2026-42345NVD/CVE Database - High
CVE-2026-42339: New API is a large language mode (LLM) gateway and artificial intelligence (AI) asset management system. In versions…
CVE-2026-42339NVD/CVE Database - Critical
CVE-2026-42302: FastGPT agent-sandbox unauthenticated remote code execution via code-server
CVE-2026-42302NVD/CVE Database - High
GHSA-pjwx-r37v-7724: LangChain vulnerable to unsafe deserialization of attacker-controlled objects through overly broad `load()` allowlists
CVE-2026-44843GitHub Advisory Database - High
GHSA-4vg5-rp28-gvjf: Open WebUI has Improper Authorization Control
CVE-2026-44567GitHub Advisory Database - Critical
GHSA-5c57-rqjx-35g2: Cline Kanban Server has a Cross-Origin WebSocket Hijacking Vulnerability
CVE-2026-44211GitHub Advisory Database - High
GHSA-gphh-9q3h-jgpp: banks has Critical Remote Code Execution (RCE) via Jinja2 SSTI
CVE-2026-44209GitHub Advisory Database - High
GHSA-8g7g-hmwm-6rv2: n8n-mcp affected by path traversal, redirect-following SSRF, and telemetry payload exposure
GitHub Advisory Database - High
GHSA-cmrh-wvq6-wm9r: n8n-mcp webhook and API client paths has an authenticated SSRF
CVE-2026-44694GitHub Advisory Database - High
CVE-2026-41487: Langfuse LLM connection update flaw exposes stored provider API key
CVE-2026-41487NVD/CVE Database - Critical
CVE-2026-42271: LiteLLM command execution through MCP test endpoints
CVE-2026-42271NVD/CVE Database - High
CVE-2026-42261: PromptHub SSRF via IPv6 private address bypass in skills fetch
CVE-2026-42261NVD/CVE Database - Critical
CVE-2026-42208: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. From version 1.81.16 to before…
CVE-2026-42208NVD/CVE Database - Critical
CVE-2026-42203: LiteLLM code execution through prompt template rendering in /prompts/test
CVE-2026-42203NVD/CVE Database - High
CVE-2026-35435: Azure AI Foundry M365 published agents improper access control
CVE-2026-35435NVD/CVE Database - High
CVE-2026-33111: Copilot Chat (Microsoft Edge) command injection allowing information disclosure
CVE-2026-33111NVD/CVE Database - High
CVE-2026-32207: Azure Machine Learning cross-site scripting allows unauthorized spoofing
CVE-2026-32207NVD/CVE Database - High
CVE-2026-26164: M365 Copilot injection flaw allows unauthorized information disclosure
CVE-2026-26164NVD/CVE Database - High
CVE-2026-26129: M365 Copilot improper neutralization enables information disclosure
CVE-2026-26129NVD/CVE Database - High
GHSA-98h9-4798-4q5v: Diffusers has a `trust_remote_code` bypass via `custom_pipeline` and local custom components
CVE-2026-44513Hugging Face Security Advisories - High
GHSA-j7w6-vpvq-j3gm: Duplicate Advisory: Diffusers has a `trust_remote_code` bypass via `custom_pipeline` and local custom components
Hugging Face Security Advisories - Critical
GHSA-9h64-2846-7x7f: Axonflow fixed bugs by implementing multi-tenant isolation and access-control hardening
GitHub Advisory Database - High
GHSA-mgx6-5cf9-rr43: Keras vulnerable to DoS via Malicious .keras Model (HDF5 Shape Bomb Causes Petabyte Allocation in KerasFileEditor)
CVE-2026-0897GitHub Advisory Database
Exploitation signals
Vulnerabilities published in the week that are listed in the CISA Known Exploited Vulnerabilities catalog or have an EPSS score of 10% or more.| Advisory | Exploitation | EPSS | Published |
|---|---|---|---|
| CVE-2026-42271: LiteLLM command execution through MCP test endpoints CVE-2026-42271NVD/CVE Database | Known exploited | 91.4% | |
| CVE-2026-42208: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. From version 1.81.16 to before… CVE-2026-42208NVD/CVE Database | Known exploited | 89.4% |
Packages that began delegating to a language model
Exposure Registry packages whose first release declaring an LLM SDK, agent framework or MCP dependency was published in the week.| Package | Ecosystem | LLM SDKs | Release | Released |
|---|---|---|---|---|
| teich | PyPI | Hugging Face Hub / Transformers | 0.1.1a1 |
Topics that moved
Largest increases over the mean of the 4 previous weeks, for topics with at least 3 records in the week.| Topic | Records | Weekly mean, previous 4 | Difference |
|---|---|---|---|
| AI agents | 19 | 12.0 | +7.0 |
| Model Context Protocol | 10 | 5.5 | +4.5 |
| Inference infrastructure | 6 | 2.3 | +3.8 |
| Coding assistants | 6 | 2.8 | +3.3 |
Research
Peer-reviewed first, then newest. Showing 8 of 17.Adaptive active-defense hardening of ML-based NIDS against RL-driven adversaries: A comparative analysis with static defenses
Peer-reviewedElsevier Security JournalsSBOMs into Agentic AIBOMs: Schema Extensions, Agentic Orchestration and Reproducibility Evaluation
Peer-reviewedACM Digital Library (TOPS, DTRAP, CSUR)QAEAS: A quantum adaptive ensemble attack system against robust deep neural networks
Peer-reviewedElsevier Security JournalsFacial image encryption algorithm based on feature recognition and chaotic scrambling
Peer-reviewedElsevier Security JournalsDeep video inpainting and video inpainting detection: A comprehensive survey from deep learning perspective
Peer-reviewedElsevier Security JournalsInterpretable Semantic Medical Image Segmentation With Style and Confidence
Peer-reviewedIEEE Xplore (Security & AI Journals)An Integrated Speech Tampering Detection Framework With Deep Neural Networks
Peer-reviewedIEEE Xplore (Security & AI Journals)Automatic Red Teaming LLM-Based Agents With Model Context Protocol Tools
Peer-reviewedIEEE Xplore (Security & AI Journals)
Policy and regulation
Newest first.Generated from the AI Sec Watch database at . Every item links to its record.