Penetration testing is legal only because the system owner has authorized it in writing, for a defined scope and under agreed rules of engagement. AI tools do not change that, but they do create new ways to break it: an autonomous agent that wanders out of scope, client data sent to an unapproved model provider, or a generated exploit run without the review a human tester would apply. This course teaches AI-assisted offensive security entirely inside an authorized engagement.
The first module is the legal and contractual foundation: written authorization from someone with authority over the systems, statements of work, scope definitions down to IP ranges, domains, accounts and cloud tenants, third-party and cloud provider testing policies, rules of engagement including testing windows, prohibited actions, stop conditions and emergency contacts, and frameworks such as TIBER-EU style threat-led testing. Educational, not legal advice. Later modules apply AI to the testing workflow: analyzing open-source reconnaissance output, prioritizing attack surface, assisting web application testing alongside tools such as Burp Suite, reviewing source code and reverse-engineering binaries with Ghidra and LLM assistance, and mapping findings to MITRE ATT&CK. A dedicated module covers autonomous and agentic testing tools: how to constrain them with scope enforcement, action allowlists and human approval, and how to log everything they do. The course closes with AI-assisted reporting that keeps findings accurate and client data protected. It does not teach attacks against systems you are not authorized to test.