Vulnerability Scanning Tools for AI Agents logopeterindia.net
Knowledge Hub

Vulnerability Scanning Tools for AI Agents

Tools that test AI agents for weaknesses such as prompt injection, tool abuse and unsafe MCP servers, plus agentic tools that scan code for vulnerabilities.

51 topics•7 categories•Updated October 2026

Two Meanings of the Topic

3 entries

The phrase covers two different jobs. The first is scanning AI agents themselves: probing models and agents for prompt injection, data leakage, jailbreaks and unsafe tool use, and checking the MCP servers, skills and model files they depend on. The second is using AI agents as scanners that find, reproduce and patch vulnerabilities in other software. This page lists both. Agents differ from ordinary applications because they act on untrusted text, call tools with real credentials and change behaviour between runs, so a one-off scan is not enough; results should feed continuous testing.

01OWASP Top 10 for LLM ApplicationsOWASP's list of the main risks for LLM applications, a common checklist for scan coverage. 02MITRE ATLASKnowledge base of adversary tactics and techniques against AI systems. 03NIST AI Risk Management FrameworkNIST framework for managing AI risk.

LLM & Agent Red-Teaming Scanners

11 entries

Automated probing of models and agents for jailbreaks, injection and unsafe behaviour.

04garakNVIDIA's open-source LLM vulnerability scanner that runs probes for prompt injection, leakage and other failure modes. 05PyRITMicrosoft's open-source Python Risk Identification Tool for generative AI red teaming; the repository moved from Azure/PyRIT, which is now archived. 06PromptfooOpen-source tool for evaluating prompts and red-teaming LLM applications and agents. 07Promptfoo repositorySource code for Promptfoo. 08GiskardOpen-source testing and scanning for LLM applications and agents; the project is also offered as a commercial product. 09DeepTeamOpen-source framework for red teaming LLM systems from the DeepEval team. 10Microsoft AI Red Teaming AgentAzure AI Foundry capability that automates red-teaming scans using PyRIT. 11MindgardCommercial automated AI red teaming platform. 12LakeraAI security vendor with red-teaming and runtime protection products. Acquired by Check Point. 13HiddenLayerAI security platform covering model scanning and runtime defence. 14Cisco AI DefenseCisco's AI security product for validation and runtime protection.

MCP Server & Agent Skill Scanners

7 entries

Check the tools and extensions an agent loads before it trusts them.

15Cisco MCP ScannerOpen-source scanner for MCP servers and tools that looks for malicious or risky definitions. 16Cisco Skill ScannerOpen-source scanner for agent skills. 17Snyk Agent ScanScanner for AI agent configurations, MCP servers and skills, continuing the earlier mcp-scan project from Invariant Labs. 18Invariant LabsSecurity research on agent and MCP attacks; the company was acquired by Snyk. 19MCP security best practicesOfficial MCP guidance on attacks and mitigations. 20Top MCP ServersMCP servers directory on this site. 21MCP Servers for AI DevelopersMCP servers for coding agents on this site.

Model File & Supply Chain Scanners

7 entries

Detect malicious code and tampering in downloaded models and dependencies.

22ModelScanOpen-source scanner for unsafe serialized model files; Protect AI is now part of Palo Alto Networks, and the last commit was in February 2026, so check its activity. 23FicklingTrail of Bits tool for analysing Python pickle files, used for unsafe model files. 24PicklescanScanner for dangerous imports in pickle files. 25Sigstore model signingSigning and verification of ML models. 26TrivyGeneral vulnerability scanner for images, repositories and IaC that also covers agent container builds. 27Software Composition AnalysisDependency scanning tools on this site. 28Container Security ToolsContainer scanning tools on this site.

AI Agents That Find & Fix Code Vulnerabilities

10 entries

Agentic tools that scan source code, reproduce findings and propose patches. Verify their output manually.

29Mantis (Google)Apache-2.0 toolkit of security review skills for AI coding agents to find, reproduce and patch vulnerabilities; Google says its output needs manual verification. 30Mantis harness guideGoogle Cloud blog post on using Mantis. 31InfoQ: Google open-sources MantisNews report on Mantis. 32StrixOpen-source autonomous agents that act like hackers to find and validate vulnerabilities in applications. 33CAI (Cybersecurity AI)Open-source framework for AI-driven security testing from Alias Robotics; the repository was archived in August 2026, so check its status. 34PentestGPTLLM-assisted penetration testing tool from an academic project. 35ButtercupTrail of Bits' cyber reasoning system from the DARPA AI Cyber Challenge. 36XBOWCommercial autonomous penetration testing platform. 37Static Application Security Testing ToolsSAST tools on this site, including Mantis. 38Penetration Testing ToolsPen-testing tools on this site.

Runtime Guardrails & Detection

6 entries

Block attacks that scanning could not prevent.

39LLM GuardOpen-source toolkit that scans prompts and outputs for risks; the repository was archived in July 2026, so check its status. 40NVIDIA NeMo GuardrailsProgrammable guardrails for LLM applications. 41Guardrails AIFramework for validating LLM inputs and outputs. 42Llama GuardMeta's PurpleLlama project with Llama Guard and security benchmarks. 43Guardrails for Agentic AIGuardrail tools on this site. 44Significance of Guardrails for Agentic AIWhy guardrails matter, on this site.

Related Pages

7 entries

Neighbouring pages on this site.

45AI Agent Security SolutionsSecuring agents. 46AI Application Security SolutionsSecuring AI applications. 47AI Security Posture ManagementAI-SPM tools. 48DAST ToolsDynamic testing tools. 49DevSecOps ToolsSecurity in the pipeline. 50Managed Cloud Infrastructure for AI AgentsSandboxes and runtimes for agents. 51Agentic AIAgentic AI hub.

Frequently Asked Questions

What does vulnerability scanning for AI agents mean?

Testing agents for weaknesses such as prompt injection, data leakage and unsafe tool use, and checking the MCP servers, skills and model files they rely on. It can also mean using AI agents to scan other software for vulnerabilities.

How is it different from a normal vulnerability scan?

Conventional scanners match known flaws in code and packages. Agents add behavioural risks: they follow instructions hidden in content, act with real credentials and may respond differently each run, so tests must be repeated and combined with runtime controls.

Which tools should I start with?

Open-source options such as garak, PyRIT and Promptfoo cover model and agent red teaming, and MCP and skill scanners cover the tools an agent loads. Add dependency and container scanning for the code around the agent.

Can AI agents be trusted to find vulnerabilities?

They can speed up discovery, but they produce false positives and inaccurate patches. Google makes the same point about Mantis, so have a security engineer verify findings and run the tools in isolation from production.

How current is this page?

Links and project activity were checked in October 2026. This area changes quickly: projects are renamed, acquired or archived, and several tools are vendor products whose claims are the vendors' own. Sites that block automated checks are kept and may need a manual click. No scanner finds every flaw, so treat results as input to human review.