Academic system
SwarmAttacker vs PentestAgent
A three-stage pipeline that fingerprints a target, retrieves and ranks candidate CVEs and exploits with retrieval-augmented generation, then attempts them, following the PTES methodology.
- Vendor
- Northwestern, Zhejiang University and Ant Group
- Licence
- MIT
- Pricing
- Free, self-hosted. You bring your own model API key or local model.
- Checked
- September 2, 2026
| Capability |  SwarmAttacker8 of 8 | PentestAgent2 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 PentestAgent's public pages or repository on September 2, 2026, not that it is technically impossible.
Profile
What is PentestAgent?
PentestAgent is a research framework from Northwestern University, Zhejiang University and Ant Group, posted in November 2024 and published at ACM AsiaCCS 2025. It follows the PTES methodology with a pipeline of specialised agents: a reconnaissance agent that fingerprints the target, search and planning agents that use retrieval-augmented generation to pull and rank candidate CVEs and exploits from public sources, and an execution agent that attempts the chosen exploits.
On its own benchmark of scanner targets and practice machines, it reports 74.2% success with GPT-4. Targets are supplied as an IP and port with optional application and version hints, so it is CVE-informed rather than URL-only black-box, and it requires several external service tokens for its exploit lookups. Refusals are handled by role-play. The agents run recon, planning and execution as ordered scripts or containers rather than in parallel.
The MIT-licensed repository has around 130 stars and its last commit dates from December 2025. Its core loop retrieves existing exploits for a fingerprinted service, which is a different task from discovering and exploiting an unknown web app from just a URL.
In SwarmAttacker's favour
Where it stands out.
- 01Black-box discovery from a URL, where PentestAgent retrieves known CVEs for a fingerprinted service.
- 02A parallel swarm, versus PentestAgent's sequential recon-plan-execute pipeline.
- 03Runs on a ChatGPT plan with no API key and no extra service tokens.
- 04Native refusal handling rather than role-play.
In PentestAgent's favour
Where it is stronger.
- 01A retrieval-augmented CVE pipeline that is strong on known-vulnerability exploitation.
- 02A peer-reviewed AsiaCCS paper and the best score among the older academic baselines.
- 03A methodology-driven design mapped to PTES phases.
Verdict
Which one should you pick?
PentestAgent is a solid academic pipeline for exploiting known CVEs on a fingerprinted host. Pick SwarmAttacker if your task is discovering and exploiting an unknown web app from a URL, in parallel, on a ChatGPT plan.
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