AI Governance ConferenceDec. 10, 12-3pm ETSave your seat →

The AI Pen-Testing
Toolbox

AI tools and resources for pen-testers, red teamers, and bug bounty hunters.

November 3, 2026 · 12-3pm ET

AI Pen-Testing Conference

A free, three-hour online event on AI-driven pen-testing and red teaming, with the people doing the work.

See the speakers →

A market with layers

AI is changing pen-testing in two ways at once. Testers have new tools: agents that find, chain and prove vulnerabilities on their own. And there is a new thing to test: the LLM apps, chatbots and agents every company is shipping.

The market that grew up around this looks like a pile of logos. It is not. It has five layers, and each one is there for a reason.

How it fits together

Models power the agents. Agents get built into platforms. People use them, check them and sign off. The shared playbook says what everyone should test for. Click a layer to jump to it.

05The sharedplaybookWhat to test for, andthe skills to do it.Guides every layer.OWASP Top 10sMITRE ATLASAVIDCertifications04The peopleFirms and bug bounty crowds.Judgment, business logic and the signed report.03The platformsPTaaS and app testing tools, already in your workflow.Now with agents alongside human testers.02The agentsAutonomous testers, commercial and open source.Recon, exploit, prove, write it up.01The modelsFrontier and open-weight LLMs.The labs unlock them for vetted defenders.used and checked bybuilt intopowerWHAT GETS TESTEDYour appsWeb apps, APIs,cloud and networks.Your AILLM apps, chatbots,agents and models.testWHAT YOU GETProven findingswith a working exploitA reportsomeone signsFixesand a retest05The shared playbookWhat to test for. Guides every layer.04The peopleFirms and bug bounty crowds.Judgment and the signed report.03The platformsPTaaS and app testing tools,now with agents built in.02The agentsAutonomous testers, commercialand open source.01The modelsFrontier and open-weight LLMs,unlocked for vetted defenders.used and checked bybuilt intopowertestYour appsWeb apps, APIs,cloud, networks.Your AILLM apps, agentsand models.WHAT YOU GETProven findings with a working exploit,a report someone signs, and fixes.

The landscape in one map

Five layers, from the foundation up. 68 tools and resources. Click any category to jump to it.

Where to start

Pick the line that sounds like you.

You run pen-tests

Let agents take the recon and repetitive checks, and spend your time on business logic and chaining.

You need a pen-test done

For a report an auditor will accept, use a firm or PTaaS. Add an agent between tests.

You're testing your own AI app

Decide what to test for with the frameworks, then check what your current tools already cover.

You hunt bugs or want to break in

Find the AI programs and payouts, prove the skill, and get access to less restricted models.

Layer by layer

Why each layer looks the way it does, then every tool in it.

Layer 01

The models

The skills that find a vulnerability can also exploit one, so the labs keep cyber safeguards switched on by default and open them up only through vetted access programs for verified defenders. The labs are also moving up the stack: OpenAI, Google and Microsoft now hand those defenders their own security agents, not just a model. Open-weight models you can run yourself are listed with the open-source tools in the next layer.

Frontier AI Access Programs

Vetted access programs that give verified defenders and pen-testing teams frontier models with relaxed cyber safeguards.

When to use it: You're on a vetted security team and the public models refuse or water down the offensive work you need.

Anthropic CVPCyber Verification Program

Anthropic's Cyber Verification Program (CVP), now merged with Project Glasswing, reduces cyber safeguards on Claude models. Its Red Team Access tier covers authorized pen-testing firms.

OpenAI Daybreak

OpenAI's Daybreak gives verified defenders GPT Cyber models, the Codex harness, and Codex Security, and lets approved partners build them into their own products.

Google FairwindGoogle DeepMind

Google DeepMind's Fairwind Program gives vetted defenders, such as governments and critical infrastructure operators, exclusive access to Gemini 4 Argon and the CodeMender agent.

Microsoft MDASHMAI-Cyber-1-Flash

Microsoft's multi-agent vulnerability scanning system, powered by its MAI-Cyber-1-Flash model and limited to verified defenders.

Layer 02

The agents

Once models could use tools and work through long, multi-step tasks, a pen-test became something an agent could attempt. XBOW reaching the top of HackerOne's leaderboard showed agents could compete with human hunters. The startups now differ mostly in how they earn trust: a working exploit attached to every finding, independent validators, or a human in the loop.

Pen-Testing AI Agents

Next-gen agentic pen-testing tools that autonomously plan, run, and validate pen-tests. Many offer continuous, on-demand testing.

When to use it: You want testing that keeps pace with your releases. Point an agent at a staging app or API and let it find, chain and prove issues between manual tests. Compare how each one validates its findings, and keep a person reviewing what it reports.

XBOW

An autonomous offensive security platform that continuously runs agents against your applications, chaining flaws into validated exploits. Topped the HackerOne leaderboard, and uses independent validators to confirm exploits.

Horizon3.aiNodeZero

AI pen-testing that applies AI to autonomously hack a system, discover vulnerabilities, and determine if they are exploitable, with remediation recommendations. Its NodeZero platform is used by the NSA and four of the Fortune 10.

Novee

Offers a proprietary offensive AI model optimized for hacking that continuously discovers vulnerabilities, determines exploitability, and provides tailored fixes.

Simbian

An agent for conducting context-aware penetration testing and analysis with remediation guidance. Fixed, transparent per-test pricing.

Terra Security

Continuous agentic pen-testing with humans in the loop. Covers web apps, networks, and AI systems, with testing aligned to code changes.

Tenzai

An AI hacker for the enterprise. Its agents find, chain, and exploit the vulnerabilities scanners miss. Built by the team behind Guardicore, with a $75M seed round.

RedVeil

Autonomous pen-testing that involves agents in multi-step attack chains. Launches on demand and delivers audit-ready reports within hours.

CodeAnt AI

Runs black-box, white-box, and gray-box pen-tests with agents that reason across code, infrastructure, and runtime to prove what's exploitable and fix it.

MindFort

YC-backed autonomous agents that continuously pen-test live apps and APIs, attach a proof-of-concept exploit to every finding, and open pull requests with patches.

Sableby Vulnetic

Sable, by Vulnetic, can run penetration tests on web apps, APIs, internal networks, and Active Directory. Runs real exploits and backs each finding with reproducible evidence.

Xintby Theori

Offensive security tester from Theori that provides both source code and runtime penetration testing. Confirms exploitable findings and offers remediation advice.

RunSybilSybil

Sybil reasons like an attacker to continuously test code, APIs, cloud, and infrastructure, chaining exploitable paths and giving security feedback on every pull request.

TurboPentest

Self-service agentic pen-testing for web apps and APIs, sold as one-time tests. Autonomous agents work alongside 14 security tools.

Open-Source AI Pen-Testing Tools

Helpful agents, open-source projects, local LLMs, and other tools for free, self-hosted pen-testing.

When to use it: You want to experiment for free, keep everything on your own hardware, or build your own agent. Also the place to start when your data can't leave your network.

Autonomous Agents

Self-hosted agents that run pen-tests autonomously, using the model of your choice.

Strix

An open-source AI penetration testing tool to find and fix your app's vulnerabilities. The agent can run locally and works with any model of your choice.

Shannonby Keygraph

Keygraph's white-box AI pen-tester for web apps and APIs. Reads your source code, then reports only the vulnerabilities it can exploit.

PentAGI

A fully autonomous multi-agent system for complex pen-testing tasks. Self-hosted with Docker and works with multiple LLM providers.

PentestGPT

The agentic pen-testing framework published at USENIX Security 2024. Now runs autonomous pipelines on Claude Code and Codex, with a legacy mode for other providers or local models.

Deadend CLIby Stray Labs

Built by Stray Labs, this is another self-hosted, free open-source tool for AI-driven penetration testing. Scores about 81% on XBOW's validation benchmark.

Frameworks and Building Blocks

Foundations for assembling your own pen-testing agents or extending tools like Claude Code.

Kali Linuxmcp-kali-server

The standard pen-testing Linux distribution. Its mcp-kali-server package lets Claude and other MCP clients run tools like Nmap, sqlmap, and Metasploit.

Cybersecurity AICAI · Alias Robotics

Alias Robotics' framework for building your own offensive and defensive agents, with support for 300+ models. Free for research use.

HexStrike AI

An MCP server that lets agents like Claude or Copilot autonomously run 150+ security tools for pen-testing and bug bounty workflows.

Transilience

Open-source skills, agents, and commands for AI-powered penetration testing and security research with Claude Code.

LLM Red-Teaming Tools

Scanners and frameworks for testing the models and AI apps themselves.

garakby NVIDIA

NVIDIA's LLM vulnerability scanner. Like nmap for models, it probes for prompt injection, jailbreaks, data leakage, and more.

PyRITby Microsoft

Microsoft's open-source red-teaming framework for proactively finding risks in generative AI systems.

Open-Weight LLMs

Pen-testers can also run and fine-tune open-weight LLMs on their own to engage in pen-testing.

KimiOpen-weight LLM

Moonshot AI's open-weight model family. Deadend CLI ran its XBOW benchmark on Kimi K2.5.

DeepSeekOpen-weight LLM

Has open-weight models that could plug into open-source pen-testing agents.

QwenOpen-weight LLM

Alibaba Cloud's open-weight model family, with sizes small enough to run on local hardware.

Layer 03

The platforms

They already own the customers and the workflows, so they are adding agents rather than losing ground to the startups. Burp Suite now has an AI partner built in. Synack pairs its agent, Sara, with 1,500+ vetted researchers. Cobalt pairs 500+ pen-testers with an autonomous offering.

Pen-Testing Tools & Services

Penetration testing as a service (PTaaS) providers and AI-assisted platforms or scanners that pair automation with human testers and existing pen-test workflows.

When to use it: You want people and automation together: a platform to book tests, track findings and retest fixes, with human testers on the hard parts. Burp Suite is here too, as the tool most testers already use day to day.

BreachLock

A full-stack PTaaS platform that pairs agentic AI with certified human testers to expose gaps in applications, APIs, cloud, networks, and IoT.

Escape

Runs agentic workflows that conduct penetration testing, discover exploits, offer next steps, and retain memory. Covers API- and GraphQL-based testing in addition to standard application vulnerabilities.

SignalFisher

Continuous, AI-assisted security testing that pairs automated scanning with vetted ethical hackers and expert validation. Formerly Bug Bounty Switzerland.

Hadrian

Agentic offensive security platform. Atlas continuously maps and tests external exposure, while Nova runs on-demand agentic pen-tests that return verified, evidence-backed findings.

Burp Suiteby PortSwigger

The pen-tester's staple web testing toolkit. Burp AI adds an agentic partner in Repeater that suggests attack angles, validates exploits, and drafts reports.

Cobalt

PTaaS provider that pairs 500+ vetted pen-testers with an autonomous pen-test offering. Also runs dedicated AI and LLM pen-tests.

StealthNet AI

Hybrid pen-testing that combines autonomous AI agents with senior US-based testers, delivering audit-ready reports in 48 hours. Also tests AI and LLM systems.

Astra

Combines pen-testing with API security and cloud vulnerability scanning. Offers features like human verification for pen-testing, AI-powered threat modeling, and SOC 2, ISO 27001, HIPAA, and PCI reports.

SynackNetSPI

PTaaS that pairs Sara, its autonomous red agent, with 1,500+ vetted researchers. FedRAMP Moderate authorized, with AI and LLM pen-testing.

Dreadnode

Infrastructure for running your own offensive AI agents, with ready-made capabilities for AI red teaming, web app pen-testing, and network operations.

Traditional App Testing Platforms

Popular traditional application testing platforms, such as DAST and SAST, that have added AI capabilities or can be used to pen-test AI-based applications, agents, or LLMs for vulnerabilities.

When to use it: You already run DAST or SAST and want to know whether it covers your AI features before you buy something new. A sensible first stop for teams testing their own LLM apps.

Aikido

Popular application security platform that provides an autonomous penetration testing tool, which simulates attacks with hundreds of AI agents. Can quickly produce SOC 2 and ISO 27001-ready reports.

Semgrep

Popular SAST platform that blends deterministic static analysis with AI to detect logic flaws like broken authorization, triage findings, and generate fixes.

Invicti

DAST-first AppSec platform with proof-based scanning for apps and APIs. Now adds agentic pen-testing and AI-driven DAST alongside SAST, SCA, and ASPM.

Corgea

AI-native SAST that catches business logic and authorization flaws and generates fixes. Has autonomous AI pen-testing with auditor-ready reports.

Layer 04

The people

An auditor or a customer wants a report from someone accountable, and the most creative attacks on business logic still come from people. The bounty platforms have also become the scoreboard where agents and humans compete, and AI now has bounty programs of its own: huntr and 0DIN pay for jailbreaks, prompt injections and model exploits.

Offensive Pen-Testing Firms

Firms, consultancies, or key individuals-for-hire that can perform modern AI-aware pen-testing. Often used for third-party compliance checks or by resellers.

When to use it: You need a signed report for an auditor, a customer or a regulator, or the target is too sensitive or complex to hand to an agent alone.

Oneleet

Provides human pen-testing, code scanning, and a virtual CISO as part of its compliance platform. Can perform tests to quickly support SOC 2 and other compliance frameworks.

Maltek Solutions

Syracuse-based offensive security firm led by practicing operators. Offers AI-augmented, human-validated web app assessments alongside network pen-testing and vCISO services.

MSP Pentesting

A white-labeled solution for MSPs and resellers that offers penetration testing services. Can conduct AI pen-testing with human-validated findings across networks, web apps, and cloud.

Bishop Fox

Veteran offensive security firm whose AI/LLM security assessments pressure-test guardrails, prompt injection, and model behavior. Also runs AI-focused red team engagements.

NetSPI

Enterprise pen-testing firm with dedicated AI/ML services, including LLM application pen-testing, jailbreak benchmarking, and continuous AI penetration testing.

AI Bug Bounty Communities

The bug bounty platforms that aggregate bug bounties from companies. Bounty hunters and pen-testers should be following these, and companies should consider listing their programs on these sites.

When to use it: Hunters: this is where the AI programs and payouts are. Companies: list your AI app or model here to get many testers on it at once.

Bugcrowd

Platform for posting bug bounty programs and crowdsourcing detection. Also offers PTaaS, AI pen-tests, and AI bias assessments.

HackerOne

Leading bug bounty platform. Also pairs AI agents with human experts for agentic pen-testing, and offers AI red teaming for models and apps.

huntr

A bug bounty platform for AI, owned by Palo Alto Networks. Runs prize-based AI exploitation challenges and a model file format vulnerability program.

0DIN

Mozilla-backed GenAI bug bounty program. Pays researchers for verified jailbreaks, prompt injections, and other exploits against frontier models and AI agents.

Intigriti

European bug bounty and PTaaS platform with 150,000+ vetted researchers, including AI specialists. A strong fit for EU data sovereignty.

Layer 05

The shared playbook

AI attacks are new enough that there is no settled playbook yet. These lists are where the field is agreeing on one, and the new certifications are where testers prove they know it.

AI-Related CVE Lists & Frameworks

The vulnerability databases and risk frameworks covering AI-related exploits that red teams should be following to stay informed on threats.

When to use it: You're scoping a test or writing up results. Use these to decide what to test for, and to map findings to names your stakeholders already know.

AVIDAI Vulnerability Database

The AI Vulnerability Database, an open-source knowledge base of failure modes across AI models, tools, and agents, with reproducible evidence.

MITRE ATLAS

MITRE's knowledge base of adversary tactics and techniques against AI systems, modeled on ATT&CK and grounded in real-world case studies.

OWASP Top 10 for LLMs and Generative AI

The standard list of risks for LLM apps, from prompt injection to excessive agency.

OWASP Top 10 for Agentic Applications

Released in December 2025, this list covers the most critical security risks facing autonomous and agentic AI systems.

OWASP API Security Top 10

The go-to list of API risks, led by broken object-level authorization. Agents hit APIs directly, so it still applies.

OWASP MCP Top 10

A beta OWASP project ranking MCP risks such as token mismanagement, tool poisoning, and shadow MCP servers.

Certifications and Trainings

Hands-on certifications, courses, and communities for building and proving AI pen-testing and red-teaming skills.

When to use it: You're building AI red-teaming skills, or need to prove them to an employer or client.

AI Security University

Free courses, including LLM Security Fundamentals and MCP Security Fundamentals, that teach the OWASP Top 10 lists through real-world attacks.

Wraith Certified AI PentesterWCAP

A hands-on, 10-scenario exam covering prompt injection, tool abuse, data exfiltration, and guardrail bypass. Free during its launch phase.

OffSec AI Red TeamerOSAI+

OffSec's AI-300 course and 24-hour proctored exam on attacking LLMs, multi-agent systems, RAG pipelines, and AI infrastructure.

HTB Certified Offensive AI ExpertHTB COAE

Hack The Box certification built on its AI Red Teamer path, developed with Google. Ends in a seven-day practical assessment and report.

SANS SEC536 and GIAC GAIPT

A three-day SANS course on attacking LLMs, RAG pipelines, agents, and MCP servers. Pairs with the GIAC AI Penetration Tester certification.

Practical AI Pentest AssociatePAPA · TCM Security

TCM Security's practical exam: two days to pen-test an agentic AI application, then two days to write a professional report.

Red Team Village

Nonprofit offensive security community behind the DEF CON village. Runs workshops, CTFs, and live online training for red teamers.

Where it's heading

The layers are blurring. The incumbents are adding agents, the agent startups are adding humans, and the labs are shipping security agents of their own. Nearly everyone is ending up in the same place: agents for speed and coverage, people for scope, business logic and judgment. Expect these categories to merge.

Two kinds of target

Most of this landscape uses AI as the tester, and the target can be any app. When the target is AI itself (your LLM app, your agent, your model), look for the pieces built for it: the OWASP and MITRE frameworks, red-teaming tools like garak and PyRIT, AI bounty programs like huntr and 0DIN, and the AI assessments from firms like Bishop Fox and NetSPI.

Missing a tool?

If you build or use an AI pen-testing tool that belongs in the Toolbox, we'd like to hear from you.

Suggest a tool →