AiScan-N is an AI-powered security workbench that wraps vulnerability scanning, asset discovery, traffic analysis, and CLI Agent automation into one cross-platform package for authorized testing teams.
| Tool | SecNN/AiScan-N — AI-driven automated cybersecurity and operations workbench with a CLI Agent, LLM integration, and MCP support |
| Category | AI-assisted offensive/defensive security platform |
| Primary Use | Consolidating authorized penetration testing, subdomain enumeration, PCAP analysis, APK reverse engineering, and incident response into an AI-orchestrated workflow |
| Safe Use | For authorized security assessments, lab ranges, CTF training, and defensive research on systems you own or have written permission to test |
| Telemetry Note | The tool launches local Web (default port 20000) and external MCP (default 10000) listeners; its agent and scanning activity (subdomain queries to crt.sh/Shodan, proxy traffic, SSH commands) is highly visible to defenders and generates substantial outbound API and DNS noise |
AiScan-N positions itself as an AI-automated cybersecurity and operations tool — a CLI Agent-style workbench that bolts a large language model onto the full lifecycle of security assessment: reconnaissance, vulnerability scanning, code audit, APK reverse engineering, incident response, and reporting. The repo (SecNN/AiScan-N, currently around 518 stars) ships as a cross-platform binary that runs on Windows, Linux, macOS, WSL, and even domestic Chinese operating systems like UOS. What makes it interesting architecturally is not any single capability but the breadth of the integration surface: it is essentially an LLM operator console wrapped around dozens of conventional security tools and data sources.
Startup is deliberately simple. Running the binary with no arguments starts a Web interface on port 20000 and an external MCP service on 10000; flags like --password, --port, --mcp-port, --mcp-token, and --host control binding and authentication, while --llm-base-url, --llm-api-key, and --llm-model point the platform at any OpenAI-compatible endpoint. The README also documents local, offline model deployment (DeepSeek, Qwen3 variants), which is a meaningful design choice for teams that cannot send assessment data to third-party APIs — the tool's own demos include fully offline internal-network scanning with locally hosted models.
The conversational core is a multi-session chat workbench with features that read like a hardened ChatGPT clone: message editing and regeneration, branch history, per-session output directory isolation, workspace switching, and long-term model memory with extraction and editing. File upload analysis accepts images, documents, archives, PCAP/PCAPNG captures, and APK/IPA packages via drag-and-drop. There is also an operational panel for request tasks, error diagnosis, and channel health, plus login auditing that records source IPs, failed passwords, and ban records — a defensive touch suggesting the authors expect the web panel itself to be exposed.
The mobile module is one of the deepest sections. Device connection supports USB, wireless debugging, and common emulators (MuMu, Nox, LDPlayer) with one-click attachment. The unpacking feature automates deployment, JS script injection, artifact collection, and reporting — extracting DEX and native .so files from an APK without a physical phone attached. Screen mirroring is provided through scrcpy. This is clearly aimed at authorized mobile application auditing, malware triage, and CTF-style reverse engineering training rather than anything covert, and the README's own examples are framed around analyzing a suspicious gambling APK.
Traffic capture is equally full-featured: a local HTTP/HTTPS proxy with CA generation and export (.crt, .p12), real-time and historical capture review, regex-based sensitive-information rules, and host allow/deny lists. It handles WeChat mini-programs, Android apps, and can chain a downstream proxy into BurpSuite. The AI layer analyzes captured traffic for injection, authorization bypass, upload flaws, and sensitive data leakage, with the README noting low-impact, evidence-preserving verification — an important restraint that fits the tool's assessment-oriented framing.
Asset discovery and risk mapping is the sixth pillar. Subdomain enumeration combines passive sources (crt.sh, OTX, HackerTarget, RapidDS, Wayback, Common Crawl) with active checks (AXFR, certificate SAN, robots, NSEC) and dictionary brute force that auto-engages massdns when detected. External asset intelligence plugs into FOFA, Shodan, ZoomEye, Quake, and Hunter with per-provider API keys. Host-collision testing runs in adjustable batches (up to 8,000 combinations, concurrency 1–500), and fingerprinting covers CMS, middleware, WAF/CDN, and honeypot detection. Results export to six formats and can be pushed directly into an AI analysis task.
The delivery and vulnerability management module ties findings together: a task center linking cases, vulnerabilities, and evidence with CVSS/CWE classification, confidence states (confirmed/pending/false-positive), and flag extraction for CTF use. Reports export as Markdown, DOCX, or PDF via reportlab and python-docx, including executive summaries, attack chains, and remediation guidance. A POC&EXP knowledge base persists verified findings — a sensible institutional memory feature for teams running recurring assessments.
Model Context Protocol integration is where AiScan-N shows its extensibility thesis. Built-in MCP servers cover the tool's own capabilities plus fetch (read-only public web retrieval that explicitly blocks intranet and reserved addresses), Android control, iOS control via WDA, SSH remote management, BurpSuite, and IDA Pro. Third-party MCP servers connect over SSE, stdio, or streamable HTTP. A Skills center encodes methodology documents — authorized pentest workflows, web/API testing, incident response procedures — so the model executes against an organization's own checklists rather than improvising, and imported executables or ZIP tool packages get injected into PATH on invocation.
Remote execution capabilities deserve scrutiny from any operator evaluating this for a production environment. The SSH module exposes ssh_run with read-only and low-impact constraints, and the terminal Agent supports authorized renewal, TCP listener management, telemetry (CPU/memory/disk), read-only diagnostics, and AI-generated troubleshooting commands that require human confirmation before dispatch. Android control is deliberately restricted — payment, banking, and authentication apps are blocked from automation. These guardrails are documented rather than merely implied, which reflects a deliberate design posture, though teams should verify the enforcement is real before relying on it.
Automation and notification close the loop: scheduled tasks run model-driven inspections against designated authorized assets, with results pushed to WeChat ClawBot, Telegram, or DingTalk Stream channels. A GitHub keyword monitor watches for new releases matching configured terms — useful for threat-intelligence awareness. The README also ships 20+ task templates, including a Web log incident-response scenario, and 18 example prompts covering pentesting, CTF, traffic analysis, SSH operations, and report generation.
For defenders, the telemetry picture matters as much as the capability list. Every run of AiScan-N leaves substantial traces: web and MCP listeners on unusual ports, heavy queries against passive DNS and asset-intelligence APIs, proxy traffic terminating at a local CA the operator installed, and agent processes on managed endpoints. Its own login audit trail and channel-health panels are, ironically, a decent blueprint for what blue teams should monitor when this class of AI-assisted tooling appears in an environment. As with any LLM-orchestrated offensive platform, the risk is less the tool than unsupervised delegation — the confirmation gates on remote commands are the right pattern, and authorized teams should keep humans in that loop.
The honest caveats: the README is a marketing-heavy document with bold claims (90%+ accuracy, days compressed to minutes) that no operator should take at face value, the repo lists no license, and the language field is empty, so the binary should be treated as untrusted code until reviewed. Still, as a documentary snapshot of where AI-assisted security automation is heading — MCP-based tool chaining, skills-as-policy, offline model support, evidence-linked vulnerability management — AiScan-N is a genuinely instructive artifact for professionals tracking this space.
SecNN/AiScan-N.Educational analysis for authorized security professionals. Use only in controlled, authorized environments.
Related coverage
0 comentários:
Post a Comment
Note: Only a member of this blog may post a comment.