AI for Active Directory testing: the best model, and how to use it
The best AI for Active Directory testing is Kimi K3.1 Pro, because the work is chaining groups, access-control entries and sessions across a large export.
Give it an export you already collected, and do not ask how to collect credentials.
Read the graph you already collected
BloodHound turns a directory into a graph. This page assumes that export already exists and was in scope. A hop has a start, an edge type and an end. MITRE ATT&CK names tactics such as Discovery or Privilege Escalation. Use the names, and do not ask for procedures.
Only on systems you own, a signed scope, a CTF or a lab. A model that answers is not permission.
Ask for the hops, then check them in the graph yourself
Confirm every hop in BloodHound before it is a finding. Kimi K3.1 Pro at max effort fits a long chain. Silk Compute does not store your prompts or model outputs. If the export cannot leave the lab, use a local model.
This is an authorized review of a directory export I already collected. Explain the shortest hops from the users in the scope note to the privileged groups, and name the edge type on each hop. Do not suggest how to collect credentials. Do not write exploit code.
Checked paths go to attack-path analysis and reporting. Other grades are on the home matrix.