Listen to the article
The self-hosted AI security platform says it solved 101 of 104 XBOW challenges, though the result is developer-reported and the benchmark is no longer considered a current measure of offensive performance.
RedAmon is an open-source, self-hosted platform designed to automate parts of penetration testing with AI agents. Its developers say it maps an organisation’s attack surface as a connected graph, uses an agent to guide reconnaissance and testing, and can prepare pull requests to help address identified weaknesses. The system is intended for security teams assessing infrastructure they own or are authorised to test.
According to RedAmon’s documentation, testing runs through an isolated Kali environment, while operators can set engagement rules covering targets, tools, scan rates and operating hours. The platform also offers human approval gates and role-based controls. Organisations that need to keep data within their own environments can use local models, including Ollama, vLLM and LM Studio, rather than relying solely on external AI services. These capabilities are described by the project itself, not independently certified.
The project reports that RedAmon solved 101 of the 104 challenges in XBOW’s web-security validation benchmark, a success rate of 97.1 per cent. RedAmon says the agent was given target URLs without source code, credentials or hints, and that successful runs are accompanied by session transcripts and walkthroughs. Its scorecard provides challenge-level entries across vulnerability types including SQL injection, server-side request forgery and path traversal. The results remain the developer’s account of its own evaluation; published transcripts and test details offer material for scrutiny, but do not amount to independent verification.
XBOW describes its benchmark as a set of practical challenges intended to help compare security tools, with 104 web-application tests spanning vulnerabilities such as SQL injection, insecure direct object references and server-side request forgery. The company has also cautioned that the benchmarks, published in 2024, are now outdated as a measure of offensive performance. RedAmon’s score therefore indicates how the system performed on that particular challenge set, rather than establishing how it would fare across live networks or every type of security assessment.
RedAmon’s documentation presents the agent as one part of a broader workflow: operators can launch a session from the main interface or from a specific node in the attack-surface graph, giving the agent context about that target and its relationships. The project says findings can inform later actions, while remediation work may be packaged as pull requests. It warns that the tool is for authorised testing, education and research, and that users must not probe systems without ownership or explicit written permission.
Source Reference Map
Inspired by headline at: [1]
Sources by paragraph:
Source: Fuse Wire Services


