Critical CVEs

New CVEs as they are published, with CVSS score, CISA KEV status, public exploits and the vendor's fix. Worth tracking, but only a small part of what Panop does: mapping what you expose to the internet, testing what an attacker can actually exploit, ranking it by business impact, and helping your team understand and fix it.

Book a demo Request a scan

Feed at a glance

Tracked
13360
Critical
2486
CISA KEV
125
Public PoC
437

Most common types

  • Broken Access Control 1691
  • Remote Code Execution 1609
  • Privilege Escalation 1307
  • Authentication Bypass 1121
  • Denial of Service 799

Published per day, last 14 days

21 Oct Critical Other 3 Nov

Source: public CVE records (NVD, CISA KEV) · last published 3 Nov 2026

Critical CVEs

Newest first · severity as published by the vendor or NVD · page 95 of 166

  • 9.8 /10 Critical
    CVE-2026-51867 No auth

    AgentGPT - Broken Access Control

    AgentGPT 1.0.0 contains an incorrect access control vulnerability caused by lack of ownership or membership binding in next/src/server/api/routers/agentRouter.ts, letting external attackers access data without proper authorization, exploit requires external network access.

  • 9.8 /10 Critical
    CVE-2026-51866 No auth

    DB-GPT - Unrestricted File Upload

    DB-GPT 0.7.5 and 0.8.0 contain an unrestricted file upload vulnerability caused by skill upload via /api/v1/skills/upload, letting attackers execute uploaded skills through /api/v1/chat/react-agent, exploit requires skill upload access.

  • 9.1 /10 Critical
    CVE-2026-51864 No auth

    DB-GPT - Path Traversal

    DB-GPT v0.7.5 and v0.8.0 contain a path traversal caused by improper validation in python_file_upload, letting remote attackers write files outside intended workspace, exploit requires no special privileges.

  • 9.1 /10 Critical
    CVE-2026-51862 No auth

    DB-GPT - Path Traversal

    DB-GPT 0.8.0 contains a path traversal vulnerability caused by improper validation in skill_upload, letting remote attackers write files outside intended workspace, exploit requires crafted request.

  • 9.8 /10 Critical
    CVE-2026-51861 No auth

    Bisheng - Command Injection

    Bisheng 2.3.0, 2.4.0, and 2.4.0-beta1 contain a code injection caused by improper input handling in src/backend/bisheng/api/v1/validate.py, letting attackers execute arbitrary code remotely, exploit requires crafted input.

  • 9.1 /10 Critical
    CVE-2026-51859 No auth

    Bisheng - Path Traversal

    Bisheng 2.3.0, 2.4.0, and 2.4.0-beta1 contain a path traversal caused by improper validation in save_download_file function, letting attackers access arbitrary files, exploit requires crafted input.

  • 9.8 /10 Critical
    CVE-2026-51858 No auth

    camel-ai camel - Command Injection

    camel-ai camel 0.2.91a1, v0.2.91a2, and v0.2.91a3 contain a command injection caused by TerminalToolkit.shell_exec allowing prompt-driven shell command execution without an approval boundary, letting attackers execute arbitrary commands remotely, exploit requires no special privileges.

  • 9.8 /10 Critical
    CVE-2026-51857 No auth

    camel-ai camel - Remote Code Execution

    camel-ai camel 0.2.91a1, v0.2.91a2, and v0.2.91a3 contain a remote code execution caused by running model-produced Python code through SubprocessInterpreter without an approval boundary, letting attackers execute arbitrary code remotely, exploit requires no special privileges.

  • 9.8 /10 Critical
    CVE-2026-51856

    agentscope - Remote Code Execution

    agentscope 1.0.18, 1.0.19 contains a remote code execution caused by unprotected execute_python_code tool exposure in RealtimeAgent session, letting remote WebSocket users execute arbitrary Python code, exploit requires WebSocket access.

  • 9.8 /10 Critical
    CVE-2026-18782 PoC publicNo auth

    Trex Digital Smart Manufacturing Systems Inc. Trex MES - SQL Injection

    Trex Digital Smart Manufacturing Systems Inc. Trex MES through 2026-09-29 contains an SQL injection caused by improper neutralization of special elements in SQL commands, letting attackers execute commands via SQL injection, exploit requires crafted input.

  • 9.1 /10 Critical
    CVE-2026-103475 No auth

    yii2-starter-kit - Broken Access Control

    yii2-starter-kit through 4.2.0 contains a broken access control caused by allowedIPs set to ['*'] in debug and Gii modules, letting unauthenticated remote attackers read sensitive data and write PHP files, exploit requires no authentication.

  • 9.8 /10 Critical
    CVE-2026-103395

    LightLLM - Remote Code Execution

    LightLLM through 1.2.0 visual_only deployments contains an insecure deserialization caused by unauthenticated RPyC service with allow_pickle enabled in remote_infer_images method, letting attackers execute arbitrary code with service account privileges, exploit requires access to visual RPyC port.

  • 9.8 /10 Critical
    CVE-2026-103110 No auth

    Pexip Infinity - Remote Code Execution

    Pexip Infinity < 38.2, 39.0, 39.1, and 40.0 contain a remote code execution caused by improper input validation on Conferencing Node, letting remote attackers execute code as unprivileged users, exploit requires network access.

  • 9.0 /10 Critical
    CVE-2026-103056 Auth required

    AiSOC - Command Injection

    AiSOC < 12.0.0 contains a command injection caused by unescaped action parameters in crowdstrike_rtr.py and endpoint.py, letting authenticated users execute arbitrary commands with SYSTEM or root privileges on managed endpoints.

  • 9.9 /10 Critical
    CVE-2026-102911 Auth required

    zosmaai pi-llm-wiki - Command Injection

    zosmaai pi-llm-wiki <= 0.11.7 contains a command injection caused by manipulation of the "url" argument in mcp/index.ts component wiki_capture_source MCP tool, letting remote attackers execute arbitrary OS commands, exploit requires no special privileges.