Rapid Threat Modelling for Developers

Introduces the course structure, learning objectives, and the three-part arc: threat model foundations, attacker tactics, and AI as a modern threat surface. Describes what learners will build, do, and be able to apply by the end.

Covers the five categories of human threat actor and three non-human attack sources, each grounded in a real incident. Gives you the first question every threat model must answer: who would target this system, and what do they already have?

Defines trust boundaries and shows where they appear in real systems: the internet edge, service-to-service calls, the database read path, third-party responses, and your own dependencies. Every boundary is a point where trust must be evaluated, not assumed.

Introduces STRIDE as six questions to ask about any feature: spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege. Each category is explained through a before-and-after code example showing the vulnerability and its fix.

Shows how to structure prompts that produce useful STRIDE threat lists from an LLM, and where AI output consistently falls short without developer-supplied context. Covers four named gap categories -- business logic, ownership models, internal service architecture, and timing side-channels -- with corrected code for each.

Introduces the SecDim Threat Thinking Matrix: a grid of five actor types against six plain-language threat categories that makes STRIDE coverage visible at a glance. Two worked examples with fully filled matrices show the feature-first and attack-first strategies in practice.

Examines how attackers exploit the implicit assumptions baked into code logic -- negative amounts, skipped steps, out-of-order requests. Two worked examples (payment threshold and shopping cart flow) show how to read assumptions out of code and close them with explicit validation.

Covers threats that originate from framework behaviour (SSTI, SSRF, mass assignment) and from business processes and physical access that surround the application. The help-desk password reset path demonstrates how a technically sound digital control can be bypassed by a human process.

Explains how a small, valid input causes disproportionate resource consumption at the CPU, memory, filesystem, and database layers -- and how application-layer denial of service differs from infrastructure flooding. Introduces domain DoS: exhausting logical business resources that standard monitoring cannot observe.

Apply the SecDim Threat Thinking Matrix to Purple Car Online, a car catalogue whose recently departed manager is now a named threat actor. Run AI-assisted STRIDE analysis on the payment flow, play the attack-and-defence game in workshop or solo mode, and document the gaps between AI output and your own findings.

Builds a full threat model for an LLM-powered feature using the same actors, boundaries, and STRIDE vocabulary from earlier modules. Covers the four risk areas -- prompt injection, tool abuse, data leakage, and integrity drift -- with the attack surface map showing the three paths untrusted text can take.

Implements the five structural controls for AI-powered features: secret isolation, tool gateway design, planner/executor separation, retrieval hardening, and output constraints. Each control is built in code, with a reference architecture showing how they compose into a safe agentic AI pattern.

Put on an attacker's hat and work through two applications across five phases: reconnaissance, weaponisation, delivery, exploitation, and patch. Exploit a path traversal vulnerability and an IDOR, fix both at the root cause level, and map each finding to STRIDE categories and the threat actors who would realistically use them.