CSE Roadmap

Common tier · about 3 hours · 12 parts

Part 10 of 12 · 1 min read

Quick notes

  • Threat modelling asks four questions: what are you building, what can go wrong, what will you do about it, and did you do a good enough job?
  • Method 1: data-flow diagram plus STRIDE per element. The mechanical enumeration produces a long list; the real work is prioritising, which here meant the flows that cross from less trusted to more trusted with no control.
  • Agents add threats STRIDE does not name, above all prompt injection on any flow into a model node. MAESTRO's seven layers are a useful checklist.
  • Method 2: attack trees. Put the attacker's goal at the root, break it into AND and OR steps with probabilities and costs, and compute the cheapest attack and the cheapest set of controls that blocks every path.
  • You do not need to stop the model from being fooled if another step in the AND chain is removed. The numbers are assumptions; the value is the structure and the traceability from threat to control to test.