What changed in AI security, Sep 21 to Sep 27, 2026
Sep 21 to Sep 27, 2026 (ISO week 2026-W39). Weeks run Monday to Sunday in UTC.
232 records published, -26 on the previous week: 51 vulnerabilities (-3), 1 incident (+1), 12 research items (no change), 167 news items (-23), 1 policy item (-1).
Critical and high advisories
Vulnerability records rated critical or high, newest first. Showing 25 of 27.- Critical
CVE-2026-101065: Obot Docker quickstart exposes admin access without authentication
CVE-2026-101065NVD/CVE Database - High
CVE-2026-100585: OpenClaw permission prompt approval bypass through MCP channel bridge
CVE-2026-100585NVD/CVE Database - High
CVE-2026-84462: Zammad security filter bypass in AI Agent configuration fields
CVE-2026-84462NVD/CVE Database - High
CVE-2026-89032: BerriAI LiteLLM tenant isolation bypass in semantic cache layer
CVE-2026-89032NVD/CVE Database - High
GHSA-3cj3-hqcr-g934: Cline: Cross-Origin WebSocket Hijacking in Cline Hub Dashboard (`/browser` endpoint)
CVE-2026-59723GitHub Advisory Database - High
GHSA-mwwr-p57h-56pf: @bytebase/dbhub's read-only mode does not prevent database writes
CVE-2026-61788GitHub Advisory Database - Critical
GHSA-fm8p-53ww-hf6w: DBHub HTTP transport DNS rebinding allows unauthenticated browser-origin SQL execution
CVE-2026-61742GitHub Advisory Database - High
GHSA-g28h-2cmm-rj9x: langchain-nvidia-ai-endpoints has local file disclosure through VLM image inputs
GitHub Advisory Database - Critical
GHSA-g5f9-3xfg-p9mf: Decepticon: Role-boundary forgery via ChatML special-token literals in web crawl output composed into LLM context
CVE-2026-61732GitHub Advisory Database - High
GHSA-6v3r-4p5c-mrp5: Language Servers for AWS vulnerable to arbitrary file write
CVE-2026-12958GitHub Advisory Database - High
GHSA-xhcr-j4j9-3gh7: Language Servers for AWS Vulnerable to Arbitrary Code Execution
CVE-2026-12957GitHub Advisory Database - High
CVE-2026-95985 - Kiro IDE Allows Agentic Writes to Global Configurations While Working in Untrusted Workspaces
AWS Security Bulletins - High
CVE-2026-96804: MLflow statsmodel flavor code execution via crafted MLmodel artifact
CVE-2026-96804NVD/CVE Database - High
CVE-2026-96775: MLflow dspy flavor unsafe pickle deserialization via crafted MLmodel artifact
CVE-2026-96775NVD/CVE Database - High
CVE-2026-18875: IBM FTM for RedHat OpenShift RAG poisoning via unauthenticated upsert
CVE-2026-18875NVD/CVE Database - High
GHSA-93xw-j965-9mx3: MCP Atlassian: Arbitrary file read/exfiltration via upload_attachment missing validate_safe_path()
CVE-2026-77258GitHub Advisory Database - High
GHSA-f6pj-qv47-g96w: MCP Atlassian: Arbitrary server-local file upload to Jira/Confluence attachments via unrestricted file_path parameters
CVE-2026-77247GitHub Advisory Database - High
GHSA-p6hp-93wp-fh6p: MCP Atlassian: Path Traversal / Arbitrary File Read in confluence_upload_attachment MCP tool (incomplete fix of GHSA-xjgw-4wvw-rgm4)
CVE-2026-77262GitHub Advisory Database - High
GHSA-5mj8-gf6m-fhw8: 9Router has an Authentication Bypass in Public LLM API via Spoofable X-9r-Real-Ip Header
CVE-2026-56681GitHub Advisory Database - High
GHSA-jjhp-8crj-mppq: @roomi-fields/notebooklm-mcp has a path traversal in vault.batch tool that allows arbitrary file write outside intended vault directory
CVE-2026-61647GitHub Advisory Database - High
CVE-2026-94627: vLLM Mooncake connector flaw enables memory exhaustion
CVE-2026-94627NVD/CVE Database - High
CVE-2026-94626: vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion…
CVE-2026-94626NVD/CVE Database - High
CVE-2026-94624: vLLM denial of service in P2P KV offloading via kv_transfer_params
CVE-2026-94624NVD/CVE Database - High
CVE-2026-94623: vLLM denial of service through NIXL connector prefix caching
CVE-2026-94623NVD/CVE Database - High
CVE-2026-94622: vLLM denial of service through NIXL connector metadata handling
CVE-2026-94622NVD/CVE 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.No vulnerability published in this week is listed as exploited or has an EPSS score of 10% or more.
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 |
|---|---|---|---|---|
| genkit-a2ui | PyPI | Genkit | 0.12.0 |
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 | 39 | 20.5 | +18.5 |
| Inference infrastructure | 14 | 5.5 | +8.5 |
| Model Context Protocol | 14 | 9.0 | +5.0 |
| Adversarial machine learning | 4 | 2.0 | +2.0 |
| Frontier model safety | 4 | 3.5 | +0.5 |
Research
Peer-reviewed first, then newest. Showing 8 of 12.Auditing Artificial Intelligence Systems: A Survey of Current Frameworks, Principles and Approaches
Peer-reviewedACM Digital Library (TOPS, DTRAP, CSUR)Benchmarking post-processing methods in local differential privacy for utility and adversarial robustness
Peer-reviewedElsevier Security JournalsRobust screen-shooting watermarking via symmetric encoder-decoder coupling and style transfer-based distortion modeling
Peer-reviewedElsevier Security JournalsFlowUAP: Evading Network Intrusion Detection Systems With Type-Specific Universal Adversarial Perturbations
Peer-reviewedIEEE Xplore (Security & AI Journals)Weakly-Supervised Prompt-Guided Privacy Enhancement for Soft-Biometric Attributes in Facial Images
Peer-reviewedIEEE Xplore (Security & AI Journals)Adaptive Defense Optimization Under Intelligent Data Poisoning Attacks in Industrial Control System
Peer-reviewedIEEE Xplore (Security & AI Journals)LLMs Cannot Reliably Judge (Yet?): A Comprehensive Assessment on the Robustness of LLM-as-a-Judge
Peer-reviewedIEEE Xplore (Security & AI Journals)When the Bee Stings: CyberCom’s AI Vulnerability
Peer-reviewedAIS eLibrary (Journal of AIS, CAIS, etc.)
Policy and regulation
Newest first.Generated from the AI Sec Watch database at . Every item links to its record.