What changed in AI security, Oct 12 to Oct 18, 2026
Oct 12 to Oct 18, 2026 (ISO week 2026-W42). Weeks run Monday to Sunday in UTC.
281 records published, -84 on the previous week: 38 vulnerabilities (-62), 0 incidents (no change), 70 research items (-7), 173 news items (-14), 0 policy items (-1).
Critical and high advisories
Vulnerability records rated critical or high, newest first.- High
CVE-2026-96620: Claude Agent SDK argument injection via resume option enabling command execution
CVE-2026-96620NVD/CVE Database - High
CVE-2026-59970: Tandoor Recipes SSRF through AI provider URL in AI actions
CVE-2026-59970NVD/CVE Database - High
CVE-2026-77474: destructive_command_guard (dcg) is a command-line safety tool that blocks destructive shell commands before AI coding…
CVE-2026-77474NVD/CVE Database - High
CVE-2026-111451: Wheels CLI deploy, package and generator inputs allow shell command execution
CVE-2026-111451NVD/CVE Database - Critical
CVE-2026-111322: Archon missing authentication exposes control API to unauthenticated attackers
CVE-2026-111322NVD/CVE Database - High
CVE-2026-76066: IntentKit authorization bypass lets callers execute owner-restricted tools
CVE-2026-76066NVD/CVE Database - High
CVE-2026-104103: Ludwig archive extraction allows writes outside extraction directory
CVE-2026-104103NVD/CVE Database - High
CVE-2026-110719: Portkey AI Gateway server-side request forgery via x-portkey-custom-host header
CVE-2026-110719NVD/CVE Database - High
CVE-2026-86063: Claude Code is an agentic coding tool. Prior to 2.1.179, the Claude Code and Claude Desktop plugin installation flow…
CVE-2026-86063NVD/CVE Database - High
CVE-2026-110207: Wukong AICRM missing authorization in user and role administration APIs
CVE-2026-110207NVD/CVE Database - High
CVE-2026-110206: Roo Code command approval bypass in parseCommandLine() via command substitution
CVE-2026-110206NVD/CVE Database - High
CVE-2026-110205: Roo Code command approval bypass through parseCommandLine() command substitution
CVE-2026-110205NVD/CVE Database - High
CVE-2026-110204: Roo Code command approval bypass through parseCommandLine substitutions
CVE-2026-110204NVD/CVE Database - High
CVE-2026-110203: Roo Code command approval bypass via quote-masking regex
CVE-2026-110203NVD/CVE Database - High
CVE-2026-110202: Roo Code command approval bypass via bare >& redirect prefix
CVE-2026-110202NVD/CVE Database - High
CVE-2026-109429: Mistral Vibe command permission gate bypass via hidden command substitution
CVE-2026-109429NVD/CVE Database - High
CVE-2026-109428: Mistral Vibe permission bypass via backslash-newline command parsing
CVE-2026-109428NVD/CVE Database - High
CVE-2026-109403: conductor-oss conductor server-side request forgery in agent card fetching
CVE-2026-109403NVD/CVE Database - High
CVE-2026-96893: GitLab EE prompt injection into AI agent instructions via conversation context
CVE-2026-96893NVD/CVE Database - High
CVE-2026-84317: Nginx UI MCP config add tool skips validation, allowing dangerous directives
CVE-2026-84317NVD/CVE Database - High
CVE-2026-109331: Network-AI security control bypass in ClaudeHookBridge PreToolUse hook
CVE-2026-109331NVD/CVE Database - High
CVE-2026-108953: mac-shell-mcp OS command injection through execute_command arguments
CVE-2026-108953NVD/CVE Database - High
CVE-2026-108836: Ollama server-side request forgery in Pull API
CVE-2026-108836NVD/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.No tracked package published its first release with an LLM SDK, agent framework or MCP dependency in this week.
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 |
|---|---|---|---|
| Prompt injection and jailbreaks | 14 | 3.0 | +11.0 |
| AI agents | 38 | 32.0 | +6.0 |
| Model Context Protocol | 13 | 10.5 | +2.5 |
| Deepfakes and impersonation | 3 | 1.3 | +1.8 |
| Robotics and embodied AI | 3 | 2.5 | +0.5 |
Research
Peer-reviewed first, then newest. Showing 8 of 70.TADR: Threat-Aligned Data Refinement against label-flipping attacks in AI-powered Intrusion Detection Systems
Peer-reviewedElsevier Security JournalsBadpatches: routing-aware backdoor attacks on vision mixture-of-experts
Peer-reviewedOpenAlex (peer-reviewed AI security)From knowledge retrieval to generation: a comprehensive study of RAG's architectures, evolution, and applications
Peer-reviewedOpenAlex (peer-reviewed AI security)Sensitive area-guided adversarial attack via Transformer-based generative network
Peer-reviewedOpenAlex (peer-reviewed AI security)A unified multi-domain framework for hallucination detection and reliability evaluation in large language models
Peer-reviewedOpenAlex (peer-reviewed AI security)Topology-aware dynamic co-evolution for robust graph learning
Peer-reviewedOpenAlex (peer-reviewed AI security)Exploring Backdoor Vulnerabilities and Fairness Challenges in Generative Models: A Survey
Peer-reviewedACM Digital Library (TOPS, DTRAP, CSUR)Federated learning with update level noise perturbation and blockchain audit logging for privacy-preserving healthcare ECG classification
Peer-reviewedOpenAlex (peer-reviewed AI security)
Policy and regulation
Newest first.No regulatory or policy records were published in this week.
Generated from the AI Sec Watch database at . Every item links to its record.