Compare

SwarmAttacker vs the field.

15 AI penetration testing agents, checked against their own product pages, repositories and papers in September 2026. Hover a heading for its definition; open a tool for the head-to-head.

Tool	Free	ChatGPT plan	Custom skills	Refusal handling	Multi-agent	MIT licence	Black-box URL	Model-agnostic
SwarmAttackerOpen-source agent
StrixOpen-source agent
hackingBuddyGPTAcademic system
ShannonOpen-source agent
XBOWCommercial platform
PentestGPTOpen-source agent
PentAGIOpen-source agent
CAIOpen-source agent
HexStrike AITool server
NodeZeroCommercial platform
MAPTAAcademic system
VulnBotAcademic system
Aikido AttackCommercial platform
Pentest Swarm AIOpen-source agent
AutoPTAcademic system
PentestAgentAcademic system

yes
not found or only partly

A cross means the capability was not found, or only partly found, on the tool's public pages or repository, not that it is technically impossible. Pricing and features change, so check the sources on each comparison page.

Definitions

What each column means.

01Free: The tool can be run without paying the vendor. Model usage may still cost money.
02Runs on a ChatGPT subscription: Signs in with a ChatGPT Plus or Pro account through the Codex backend. No API key and no per-token bill.
03Custom prompts & skills: You can add your own prompts, skills or attack playbooks to steer what the agent tries, without forking the code.
04Safety-refusal handling: Detects when the model refuses an authorized request and recovers on its own: retry, rephrase, or fall back to another model.
05Parallel multi-agent swarm: A planner fans out several specialist agents that attack the target at the same time, instead of one sequential agent loop.
06MIT-licensed open source: The full product is published under the permissive MIT licence, usable in commercial work without copyleft obligations.
07Black-box from a URL: Needs only the target URL. No source code, no repository access, no credentials.
08Model-agnostic: Works with OpenAI, Anthropic, OpenRouter or local models rather than being tied to one vendor.

Head-to-head

One page per tool.

Each one lists the evidence and the public sources it was checked against.

Turn a URL into a security report.

Open source, MIT licensed, and it runs on the ChatGPT plan you already pay for. Install it and point it at a target you own.

$brew install joloooo/swarm/swarm

Source on GitHub