AI Security Agents for Pentesting, Code Fixes, and Compliance
22 tools · listed in dataset order, no ranking
The tools here do four different security jobs. Application pentesting agents attack a running web app or API you are authorized to test and report issues they can prove (Strix, Shannon, Fleuret, Hadrian Nova). Code remediation agents read source code, reproduce vulnerabilities, and draft patches (CodeMender, Codex Security, Google Mantis, Anthropic's OSS Scanner). Attack-surface management maps and checks your internet-facing assets continuously (Hadrian Atlas). Compliance automation collects audit evidence and policies for frameworks such as SOC 2 (Comp AI). They need different access, run in different places, and produce different evidence, so start with the job.
Strix
Strix (OmniSecure, Inc.)
Strix: Open-Source AI Pentesting Agents for Authorized App Security Testing
Best for: Engineering and AppSec teams that want open-source, AI-driven pentesting of their own apps and APIs in the CLI or CI, with an option to move to a managed platform
Shannon
Keygraph
Shannon: Open-Source White-Box AI Pentester for Web Apps and APIs
Best for: Engineering teams with access to both the source code and a staging copy of their web app or API who want open-source, evidence-backed security testing in CI
CodeMender
CodeMender: Google Cloud's AI Agent That Finds, Verifies, and Fixes Code Vulnerabilities
Best for: Security and platform teams already on Google Cloud who want an agent that verifies vulnerabilities and proposes tested patches, and who can join a preview program
Comp AI
Comp AI
Comp AI: Open-Source AI Compliance Platform for SOC 2, ISO 27001 and HIPAA
Best for: Startups and mid-market companies working toward SOC 2, ISO 27001, HIPAA or GDPR who want evidence collection and policy work automated, and engineering-led teams that value an open-source, self-hostable compliance platform
Codex Security
OpenAI
Codex Security: OpenAI's AI Agent for Finding, Validating, and Fixing Code Vulnerabilities
Best for: Security and engineering teams that already use Codex or the OpenAI API and want an agent that validates findings and prepares reviewable fixes, from a one-off repository audit to per-PR checks in CI
Google Mantis
Google Mantis: Open-Source Agent Skills for Finding, Reproducing, and Patching Vulnerabilities
Best for: Security engineers and AppSec teams who can operate isolated sandboxes and want a free, adaptable pipeline that reproduces and patches vulnerabilities with their own coding agent and models
Fleuret
Fleuret AI (FLEURET AI SAS)
Fleuret: EU-Hosted AI Pentesting Agents for Web Apps and APIs
Best for: European companies of roughly 50 to 1,000 people with a real web and API attack surface (SaaS, fintech, healthtech) and a NIS2, DORA, SOC 2 or ISO 27001 deadline, especially those that want pentest data and model inference kept on EU infrastructure
Hadrian
Hadrian Security B.V.
Hadrian: Agentic Attack Surface Management (Atlas) and On-Demand AI Pentesting (Nova)
Best for: CISO-led security teams at mid-market and enterprise organizations (Hadrian names 300 to 10,000 employees) with a large, changing internet-facing estate across cloud and on-premises, who want validated external exposures instead of scanner noise and pentest evidence for NIS2, DORA, ISO 27001 or SOC 2
OSS Scanner
Anthropic
OSS Scanner: Anthropic's Free, Opt-In AI Vulnerability Scanning Service for Critical Open-Source Projects
Best for: Core maintainers of established, security-critical open-source projects, such as libraries that process untrusted input, whose teams can already triage a steady flow of high and critical vulnerability reports and want them as early as possible
Apate
Apate.ai
Apate: AI Victim Personas That Engage Scammers and Turn Scam Calls and Messages into Fraud Intelligence
Best for: Fraud, scam-prevention and threat-intelligence teams at telcos, banks, crypto exchanges and government agencies that already detect scam traffic and want it engaged by AI personas to extract mule accounts, wallets, numbers and campaign intelligence, with an audit trail for regulators
Axonius AI Workforce
Axonius
Axonius AI Workforce: Virtual Workers That Triage Exposures and Act on Axonius Asset Data
Best for: Security and IT operations teams that already run Axonius on its SaaS deployment, ingest most of their tools through adapters, and want ongoing follow-up work (exposure triage, coverage gaps, ticket routing) done by agents that ask before they change anything
Purple AI
SentinelOne
Purple AI: SentinelOne's Agentic SOC Analyst That Triages, Investigates and Responds to Alerts Inside Singularity
Best for: SOC teams that run SentinelOne Singularity (or are consolidating on it, with AI SIEM for third-party data) and want alerts triaged and investigated to a verdict automatically, with response actions gated by policies they set
Material Google Drive Security Agent
Material Security
Material Google Drive Security Agent: AI Agent That Investigates Sensitive Drive Files and Fixes Oversharing in Google Workspace
Best for: Security and IT teams at Google Workspace companies with large, long-lived Drive estates who want sensitive files reviewed and permissions tightened automatically, especially ahead of rolling out Gemini or other AI tools that read Drive
Claude Security
Anthropic
Claude Security: Anthropic's AI Agent That Continuously Scans Repositories, Validates Vulnerabilities, and Tracks Fixes
Best for: Security and engineering teams at Claude Enterprise organizations that want continuously updated, validated findings across their own repositories, with severity tuned to their threat model and fixes handled in Claude Code
Neo
ProjectDiscovery
Neo by ProjectDiscovery: AI Security Agent for Autonomous Pentesting, PR Security Review and Vulnerability Triage
Best for: AppSec and product security teams, MSSPs and bug bounty hunters who want an agent to pentest running apps, review pull requests and reproduce incoming vulnerability reports with proof, on a platform that keeps context between runs
Mave
Mave
Mave: Agentic SecOps Platform That Investigates Alerts, Hunts Threats and Runs Governed Response Across Your Existing Security Tools
Best for: Enterprise SOCs, MSSPs and organizations running several SIEMs or EDRs that want AI agents to investigate every alert, hunt and tune detections across the tools they already own, with response actions under their own approval rules and no data migration
XBOW
XBOW
XBOW: Autonomous AI Pentesting for Web Apps and APIs, With a Working Exploit Behind Every Finding
Best for: Enterprise and mid-market security teams that own more web applications than they can pentest by hand and want exploit-validated findings, fix retesting and audit evidence across the whole portfolio, bought through sales or a cloud marketplace
Dropzone AI
Dropzone AI
Dropzone AI: Autonomous AI SOC Analyst and Threat Hunter That Investigate Alerts and Run Hunts Across Your Existing Security Stack
Best for: Enterprise SOCs, lean security teams that need round-the-clock alert coverage without hiring for night shifts, and MSSPs that want every alert across their SIEM, EDR, identity, cloud and email tools investigated to an evidence-backed verdict, with the team setting policy and approving containment
Decepticon
Bitter Security
Decepticon: Open-Source Autonomous Red Team Agent for Authorized Engagements
Best for: Experienced red teamers, pentesters and security researchers who want a self-hosted, open-source agent that runs full adversary-emulation chains, including post-exploitation, Active Directory and C2, against scoped targets they are authorized to attack
AiSOC
Open source (maintained by Beenu Arora, funded by Cyble)
AiSOC: Open-Source, Self-Hosted AI SOC With Auditable Triage and Investigation Agents
Best for: Security teams and MSSPs that want an AI SOC they can self-host, audit line by line and modify, including in air-gapped or regulated environments, and that have the engineering capacity to run a multi-service Docker or Kubernetes stack and choose their own models
Hacktron
Hacktron
Hacktron: AI Security Agents for White-Box Pentests, Pull Request Security Review and Continuous Red Teaming
Best for: Engineering-led SaaS companies and AppSec teams that want exploit-focused PR security review on every change plus code-aware pentests with audit-ready reports, bought self-serve with published per-seat and per-codebase-size pricing
Brava
Brava Security
Brava: Multi-Agent Detection Engineering Platform That Proves Which Attacks Your SIEM Would Catch, Fixes the Gaps and Cuts SIEM Ingestion
Best for: Enterprise detection engineering and SecOps teams running a SIEM such as Splunk, Microsoft Sentinel, Elastic or CrowdStrike NG-SIEM that need continuous, evidence-backed proof of which attacks their detections catch, want missing rules written and verified for them, and are under pressure to cut SIEM ingestion costs
How to choose an AI security agent
Running app, source code, or compliance evidence
Strix and Shannon test a running application from your machine or CI; Strix can also work black-box from a URL alone, while Shannon needs both the source code and the running app. Fleuret and Hadrian Nova run the pentest for you as a service and deliver a report. CodeMender, Codex Security, and Mantis start from the repository, try to reproduce each issue in a sandbox, and propose patches without attacking a deployment. Hadrian Atlas watches your external assets continuously. Comp AI prepares the evidence an auditor asks for; it also advertises AI pentests, but it does not perform the audit, and compliance evidence is not a substitute for security testing.
Get written authorization and pick the environment
Strix says to run it only against systems you own or have explicit written permission to test, within the agreed scope. Fleuret requires a signed authorization defining scope and testing window, and Hadrian's terms require you to own the target or hold the rights to test it. Vendors differ on where to test: Shannon says not to run it against production, Hadrian suggests staging before a launch, and Fleuret tests production with read-only proofs of concept or a staging copy if you prefer. Mantis should run only in isolated environments with no access to production systems.
Setup, cost, and who reviews the results
Strix, Shannon, and Mantis are open source and need Docker or your own infrastructure plus an LLM provider, so each run costs model tokens. Codex Security needs Codex Security access on an OpenAI plan, and CodeMender is a Google Cloud preview for selected customers with token-based pricing. Fleuret charges a flat €4,000 per web app and Hadrian Nova starts at €1,250 per test, while Atlas and Comp AI are quoted. Every vendor expects people to review findings and patches: Google says Mantis findings must be verified by a security expert, and Anthropic says OSS Scanner's fast reports arrive before human review.
Compare by job, required access, deployment, and evidence
Every security listing, using what each vendor publishes in its docs, READMEs, pricing pages, and terms, checked on October 15, 2026. We have not run these tools against a common target.
- Job
- Pentests running web apps and APIs; can also review source code
- Built for
- Developers and security teams testing their own apps
- Deployment and cost
- Open source CLI with Docker and your LLM key, or in CI; cloud Pro $29/seat plus per-pentest credits
- Access it needs
- A target URL or API spec, and optionally a repository
- Evidence and human review
- Strix says each reported finding includes a proof-of-concept exploit and reproduction steps
- Job
- White-box pentests of web apps and APIs
- Built for
- Engineering teams with a staging environment
- Deployment and cost
- Open source via npx with Docker, GitHub Actions, or GitLab CI; hosted Pro $50/developer
- Access it needs
- Source code (read-only) plus the running staging app
- Evidence and human review
- Keygraph says Shannon reports only issues it exploited (no exploit, no report)
- Job
- Managed pentests of web apps, APIs, and external infrastructure
- Built for
- EU companies needing pentest evidence for NIS2, DORA, ISO 27001, or SOC 2
- Deployment and cost
- Managed service hosted in Paris; €4,000 per web app with a re-test
- Access it needs
- Scope, test accounts per role, and a signed authorization
- Evidence and human review
- Fleuret says each reported issue comes with a replayable, non-destructive proof of concept
- Job
- Continuous attack-surface management (Atlas) and single-app pentests (Nova)
- Built for
- CISO-led teams at mid-market and enterprise companies
- Deployment and cost
- Cloud platform; Nova from €1,250 per test; Atlas quoted by asset count
- Access it needs
- Your domains for Atlas; one target URL per Nova test
- Evidence and human review
- Hadrian says Atlas confirms which exposures are exploitable; a pentest report for each Nova test
- Job
- Finds, verifies, and patches vulnerabilities in source code
- Built for
- Security and platform teams on Google Cloud
- Deployment and cost
- Google Cloud public preview for selected customers; token-based pricing
- Access it needs
- Source code you own or are authorized to use, or OSI-licensed open source
- Evidence and human review
- Tested fixes delivered as diffs for developers to review
- Job
- Threat-models repositories, validates findings, and drafts patches
- Built for
- Teams with Codex Security access on an OpenAI plan
- Deployment and cost
- Codex plugin, open-source CLI and SDK, GitHub PR review, or cloud for connected repos
- Access it needs
- Repository access; connected GitHub repos for the cloud version
- Evidence and human review
- Validates in isolation when possible; you review patches before opening a PR
- Job
- Vulnerability research pipeline: threat model, hunt, reproduce, patch
- Built for
- Security engineers adapting their own pipeline
- Deployment and cost
- Open source; your coding agent, models, and sandbox infrastructure; Google says it is for demonstration, not production use
- Access it needs
- Source code, in an isolated environment
- Evidence and human review
- Google says findings must be verified by a security expert before reporting
- Job
- Scans enrolled open-source projects and emails vulnerability reports
- Built for
- Core maintainers of established, security-critical open-source projects
- Deployment and cost
- Free, opt-in; maintainers apply and each enrollment is reviewed
- Access it needs
- Your public project's repository
- Evidence and human review
- Reports with reproducer and, where available, a candidate patch; Anthropic says its fast reports are sent before human review
- Job
- Compliance automation: evidence, policies, cloud monitoring, AI pentests
- Built for
- Startups and mid-market companies preparing for SOC 2, ISO 27001, HIPAA, or GDPR
- Deployment and cost
- Hosted platform with custom pricing; open-core code you can inspect or self-host
- Access it needs
- Cloud, identity, HR, and device integrations
- Evidence and human review
- Audit evidence and policies you approve; Comp AI says the auditor you choose issues the audit report
| Tool | Job | Built for | Deployment and cost | Access it needs | Evidence and human review |
|---|---|---|---|---|---|
| Strix | Pentests running web apps and APIs; can also review source code | Developers and security teams testing their own apps | Open source CLI with Docker and your LLM key, or in CI; cloud Pro $29/seat plus per-pentest credits | A target URL or API spec, and optionally a repository | Strix says each reported finding includes a proof-of-concept exploit and reproduction steps |
| Shannon | White-box pentests of web apps and APIs | Engineering teams with a staging environment | Open source via npx with Docker, GitHub Actions, or GitLab CI; hosted Pro $50/developer | Source code (read-only) plus the running staging app | Keygraph says Shannon reports only issues it exploited (no exploit, no report) |
| Fleuret | Managed pentests of web apps, APIs, and external infrastructure | EU companies needing pentest evidence for NIS2, DORA, ISO 27001, or SOC 2 | Managed service hosted in Paris; €4,000 per web app with a re-test | Scope, test accounts per role, and a signed authorization | Fleuret says each reported issue comes with a replayable, non-destructive proof of concept |
| Hadrian | Continuous attack-surface management (Atlas) and single-app pentests (Nova) | CISO-led teams at mid-market and enterprise companies | Cloud platform; Nova from €1,250 per test; Atlas quoted by asset count | Your domains for Atlas; one target URL per Nova test | Hadrian says Atlas confirms which exposures are exploitable; a pentest report for each Nova test |
| CodeMender | Finds, verifies, and patches vulnerabilities in source code | Security and platform teams on Google Cloud | Google Cloud public preview for selected customers; token-based pricing | Source code you own or are authorized to use, or OSI-licensed open source | Tested fixes delivered as diffs for developers to review |
| Codex Security | Threat-models repositories, validates findings, and drafts patches | Teams with Codex Security access on an OpenAI plan | Codex plugin, open-source CLI and SDK, GitHub PR review, or cloud for connected repos | Repository access; connected GitHub repos for the cloud version | Validates in isolation when possible; you review patches before opening a PR |
| Google Mantis | Vulnerability research pipeline: threat model, hunt, reproduce, patch | Security engineers adapting their own pipeline | Open source; your coding agent, models, and sandbox infrastructure; Google says it is for demonstration, not production use | Source code, in an isolated environment | Google says findings must be verified by a security expert before reporting |
| OSS Scanner | Scans enrolled open-source projects and emails vulnerability reports | Core maintainers of established, security-critical open-source projects | Free, opt-in; maintainers apply and each enrollment is reviewed | Your public project's repository | Reports with reproducer and, where available, a candidate patch; Anthropic says its fast reports are sent before human review |
| Comp AI | Compliance automation: evidence, policies, cloud monitoring, AI pentests | Startups and mid-market companies preparing for SOC 2, ISO 27001, HIPAA, or GDPR | Hosted platform with custom pricing; open-core code you can inspect or self-host | Cloud, identity, HR, and device integrations | Audit evidence and policies you approve; Comp AI says the auditor you choose issues the audit report |
Prices are the vendors' published starting figures. Editorial comparison, not a ranking or security advice.
Common tradeoffs
Substitutes vs complements
Strix, Shannon, Fleuret, and Hadrian Nova are alternatives for testing a running app. CodeMender, Codex Security, and Mantis are alternatives for source-code remediation, and they complement a pentest because they find different issues from a different starting point. Comp AI and Hadrian Atlas sit alongside both rather than replacing either.
Run it yourself vs buy a service
Open-source tools give you control over where code and findings go, but you maintain the setup and pay for model usage. Fleuret and Hadrian deliver reports without setup, at a per-test price, and suit teams that need evidence for auditors or customers.
Black-box vs white-box
Testing from a URL alone shows what an outside attacker can reach. Giving a tool your source code, as Shannon and the remediation agents require, finds more and can produce patches, but means sharing code with the tool and its model provider.
Compliance vs security testing
Compliance automation proves controls exist for an auditor. It does not show that an application is free of exploitable flaws, so companies preparing for SOC 2 or ISO 27001 still need separate testing.
Common questions about security tools
What is an AI security agent?
An AI agent that does security work itself, such as attacking an application you are authorized to test, reproducing a vulnerability in source code and drafting a fix, mapping exposed assets, or collecting compliance evidence, rather than only listing potential issues.
Are there open-source AI pentesting tools?
Yes. Strix is Apache-2.0 and Shannon is AGPL-3.0; both run locally or in CI with your own model provider, and both vendors also sell hosted platforms. Google Mantis is Apache-2.0 but focuses on source-code research and is published for demonstration rather than production use. See Strix →
Can I run an AI pentest against production?
Only with written authorization, and only if the vendor supports it. Shannon says not to run it against production systems. Fleuret tests production with read-only, non-destructive proofs of concept, or a staging copy if you prefer, and Hadrian suggests testing staging before a launch. A staging environment with disposable data is the safer default for tools you run yourself.
Does a compliance platform like Comp AI replace a pentest?
No. Comp AI collects evidence, drafts policies, and monitors cloud settings so an auditor can check your controls, and it advertises AI pentests as one feature. Comp AI says the audit report comes from the auditor you choose, and passing an audit does not mean an application has no exploitable flaws. See Comp AI →
Do AI security agents replace human pentesters?
No. Vendors say findings and patches still need human review, and coverage is narrower than a full manual assessment. They are best used for continuous or pre-release checks alongside periodic human-led testing.
Related
- Browse AI coding agents
Codex, Claude Code, and other coding agents that some security tools here plug into.
- See the best AI coding agents
For choosing the coding agent that would apply or review the patches these tools propose.
Vendor facts in this guide last checked October 15, 2026 against: Strix README, Strix pricing, Shannon README, Keygraph pricing, Fleuret FAQ, Hadrian pricing, Hadrian Nova terms, CodeMender, CodeMender docs, Codex Security docs, Google Mantis README, OSS Scanner, Comp AI README, Comp AI pricing. Picks and tradeoffs are editorial judgments.