AI-Driven Automated Penetration Testing Platform

You are the lead architect of an AI-driven automated penetration testing platform. Guide me through designing and building a system where autonomous security agents perform end-to-end penetration testing: 1. Agent Architecture: Design a multi-agent system where specialized agents handle reconnaissance (OSINT, port scanning, service fingerprinting), vulnerability scanning (CVE matching, configuration auditing), exploitation (payload generation, privilege escalation), post-exploitation (lateral movement, data collection), and reporting. Define inter-agent communication, task delegation, and a shared knowledge graph. 2. AI Core Capabilities: Implement an LLM-based reasoning engine that selects attack techniques based on target context, generates custom payloads adapted to detected technologies, and decides exploitation order using reinforcement learning trained on historical engagements and MITRE ATT&CK. 3. Target Ingestion & Scope: Define how targets ([INSERT_IP_RANGES], [INSERT_DOMAINS], [INSERT_CLOUD_ACCOUNTS]) are ingested, scope boundaries enforced (allow/deny lists), and rules of engagement checked automatically before any action. 4. Tool Orchestration: Integrate tools (Nmap, Nuclei, Burp Suite API, Metasploit, custom scanners) behind a unified abstraction layer, with agents selecting and chaining tools dynamically. Build a sandbox for safe payload testing. 5. Safety & Guardrails: Implement rate limiting, dry-run mode, human-in-the-loop approval gates for destructive actions, and a kill switch. Ensure no action violates [INSERT_ROE] and the platform cannot pivot to out-of-scope assets.