Academic system
SwarmAttacker vs VulnBot
A tri-phase multi-agent pentest framework that hands work between reconnaissance, scanning and exploitation agents via a task graph, targeting IP-based machines from a Kali shell.
- Vendor
- Institute of Information Engineering, Chinese Academy of Sciences
- Licence
- MIT
- Pricing
- Free, self-hosted. You bring your own model API key or local model.
- Checked
- September 2, 2026
| Capability |  SwarmAttacker8 of 8 | VulnBot2 of 8 |
|---|---|---|
| Free | Yes. | |
| Runs on a ChatGPT subscription | Not found. | |
| Custom prompts & skills | Partial. | |
| Safety-refusal handling | Partial. | |
| Parallel multi-agent swarm | Partial. | |
| MIT-licensed open source | Yes. | |
| Black-box from a URL | Not found. | |
| Model-agnostic | Partial. |
- yes
- not found or only partly
Hover a mark for the evidence. A cross means the capability was not found, or only partly found, on VulnBot's public pages or repository on September 2, 2026, not that it is technically impossible.
Profile
What is VulnBot?
VulnBot is a research prototype from the Institute of Information Engineering at the Chinese Academy of Sciences, published in January 2025. It models a pentest team as three sequential role agents, a Collector for reconnaissance, a Scanner, and an Exploiter, coordinated by a penetration task graph and a summariser that passes condensed findings between phases. Commands run over SSH on a Kali host, state lives in a database, and an optional retrieval module supplies knowledge.
The agents run one role at a time rather than in parallel. Targets are described in free text and must include an IP or port, so it is built for IP-based practice machines rather than URL-only web testing. Refusals are handled by a role-play prompt asserting the test is authorized and simulated. The MIT-licensed code, around 3,000 lines of Python with roughly 190 stars, has had no commits since April 2025.
VulnBot's own paper reports results on an open-source privilege-escalation benchmark and 13 practice machines. The authors note that end-to-end testing of real machines remains a significant challenge and that exploitation is its weakest phase.
In SwarmAttacker's favour
Where it stands out.
- 01A parallel swarm versus VulnBot's strictly sequential role hand-off.
- 02Black-box from a URL, where VulnBot needs an IP-based target and a Kali host.
- 03Runs on a ChatGPT plan and handles refusals, where VulnBot uses a role-play jailbreak and an API key.
- 04Actively maintained, where VulnBot has been dormant since April 2025.
In VulnBot's favour
Where it is stronger.
- 01A published task-graph design for coordinating recon, scanning and exploitation phases.
- 02A focus on IP-based practice machines and privilege escalation that a web-only tool does not cover.
- 03An academic paper with reproducible benchmarks on open-weight models.
Verdict
Which one should you pick?
VulnBot is a research prototype worth reading for its task-graph design, but it is dormant and built for practice machines. Pick SwarmAttacker for a maintained, parallel, black-box web agent.
Sources
What this page was checked against.
All checked on September 2, 2026.
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