Pentest Swarm AI Tool Exposes Critical Vulnerabilities in Nmap, SQLMap, Burp, Metasploit

Key Takeaways

• Armur AI has launched Pentest Swarm AI, an open-source autonomous penetration testing platform utilizing swarm intelligence.
• Unlike traditional multi-agent systems, it employs stigmergy, emergence, and decentralization for dynamic attack chain formation.
• The platform integrates popular offensive security tools like Nmap, SQLMap, Burp Suite, Metasploit, and ProjectDiscovery tools.
• It supports various AI models, including Claude and Ollama, and generates submission-ready reports with automated CVSS scoring.
• Pentest Swarm AI is available under an AGPL-3.0 license, promoting community contributions from commercial forks.

Armur AI has unveiled Pentest Swarm AI, an innovative open-source platform designed for autonomous penetration testing. This new tool distinguishes itself by leveraging swarm intelligence principles rather than the conventional, sequential execution typical of most multi-agent security tools.

The platform offers cybersecurity professionals a dynamic and coordinated interface to a comprehensive suite of offensive tools, including Nmap, SQLMap, Burp Suite, ZAP, and Metasploit. These capabilities are driven by a user-selected AI model, providing flexibility and power to security operations.

Understanding True Swarm Intelligence

Many tools marketed as “multi-agent” are essentially predefined pipelines, where a central planner dictates a fixed sequence of actions such as reconnaissance, classification, exploitation, and reporting. Pentest Swarm AI breaks this paradigm by implementing three core swarm-intelligence primitives:

• Stigmergy: Agents coordinate indirectly by writing findings to and reading from a shared PostgreSQL-backed blackboard (pgvector). Each finding is assigned a “pheromone weight” that influences other agents to prioritize high-value attack paths, with weights naturally decaying over time to sideline stale paths.
• Emergence: Attack chains are not pre-scripted but form organically. For instance, a reconnaissance finding can activate a classifier agent, which, upon identifying a high-severity CVE, can trigger an exploit agent. The results of the exploit then cycle back into the blackboard, fostering continuous adaptation.
• Decentralization: Every agent operates based on its own trigger predicates, meaning that adding or removing agents does not necessitate a rewrite of a central orchestrator.

The platform currently supports eight ProjectDiscovery tools out of the box: subfinder, httpx, nuclei, naabu, katana, dnsx, and gau. It also includes a fully parsed Nmap XML adapter with integrated scope validation. Future updates (Wave 2) are slated to include adapters for SQLMap, Burp MCP bridge, Metasploit, and ZAP, enhancing its capabilities without requiring a fundamental platform redesign.

Initiating a scan with Pentest Swarm AI is straightforward, requiring only an API key and a single command:

export PENTESTSWARM_ORCHESTRATOR_API_KEY=sk-ant-your-key-here
pentestswarm scan example.com --scope example.com --swarm --follow

The system offers flexibility in AI model selection, supporting Claude (the default, with prompt caching for recon and classifier agents), Ollama for secure, air-gapped local deployments, and any OpenAI-compatible model. This allows teams to balance cost-efficiency, privacy concerns, and functional requirements. Cloud-based operations do not require a GPU or local model downloads.

Each campaign generates comprehensive, submission-ready reports in Markdown, HTML, JSON, and SARIF formats, all directly queried from the blackboard by a dedicated reporting agent. Findings are automatically deduplicated, scored according to the FIRST CVSS v3.1 specification, and rigorously scoped. The --scope flag enforces boundaries at both the tool and executor layers, providing defense-in-depth suitable for CI/CD pipelines and bug bounty programs.

GitHub Actions & MCP Integration

Pentest Swarm AI includes a pre-built GitHub Action that outputs SARIF, facilitating automated penetration testing directly within CI/CD workflows. Furthermore, the pentestswarm mcp serve command exposes the entire swarm as an MCP server, enabling native integration with tools like Claude Desktop and Cursor for enhanced IDE-level offensive security testing.

The project is licensed under AGPL-3.0, making it freely available for red teams, bug bounty hunters, and internal security pipelines. This copyleft license ensures that any commercial Software-as-a-Service (SaaS) fork must contribute its improvements back to the open-source community. The platform is available on GitHub.

What You Should Do

• Explore the Platform: Red teams and security professionals should investigate Pentest Swarm AI’s capabilities for autonomous penetration testing.
• Integrate with CI/CD: Consider integrating the GitHub Action with SARIF output into your CI/CD pipelines for continuous security testing.
• Leverage AI Model Flexibility: Choose an AI model (Claude, Ollama, OpenAI-compatible) that best fits your organization’s cost, privacy, and capability requirements.
• Review Reporting: Utilize the automated, standards-compliant reporting features to streamline vulnerability management and compliance.
• Contribute to Open Source: If developing commercial forks, adhere to the AGPL-3.0 license by contributing improvements back to the community.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.