The Adversarial Frontier
Poisoning, prompt injection, manipulated inputs
The security module. It covers how models are attacked rather than merely misused — adversarial evasion, training-data poisoning, and prompt injection against AI-integrated applications — and connects each to the concrete financial threat of manipulated inputs against fraud and underwriting models. It closes the loop with Module 5 by showing how a manipulated application defeats an automated decision, and it sets a realistic defensive posture: what can be hardened, what cannot, and where a human must remain in the loop.
- Explain evasion, poisoning and prompt-injection attacks in plain terms and where each strikes a financial workflow.
- Recognise manipulated or adversarial inputs aimed at fraud, credit and monitoring models.
- Specify a defensive posture and the limits of defence, and know when to keep a human in the loop.
- Adversarial-threat literacy
- Input-integrity checking
- Recognising indirect prompt injection in AI-assisted tools
- Setting realistic expectations of model robustness
Lessons in this module
How it lands across the four desks
Your models are direct targets: evasion attacks tuned to slip past detection, poisoning that corrupts what the model learns, and manipulated inputs engineered to score clean. You learn to recognise the attack signatures and the limits of your defences.
This module closes Module 5's loop: how a manipulated application — adversarially tuned inputs, gamed features — defeats an automated credit decision, and why input integrity is now part of underwriting.
You govern the controls and set robustness expectations: what a vendor can honestly promise, what 'adversarially robust' does and does not mean, and where human-in-the-loop is a security requirement rather than a preference.
The AI assistants you adopt are attack surface too: indirect prompt injection through documents and web content you ask a tool to process. You learn the risk before you delegate a task to an agent that reads the open internet.
Key literature · 7 sources
Every module rests on a verified scholarly and institutional evidence base. The full core and further reading lists open with the module.
- Szegedy, C. et al. (2014) 'Intriguing Properties of Neural Networks.' ICLR 2014 — the discovery of adversarial examples.
- Goodfellow, I. J., Shlens, J. & Szegedy, C. (2015) 'Explaining and Harnessing Adversarial Examples.' ICLR 2015 — the mechanism and the reliability/transferability of evasion.
- Biggio, B. & Roli, F. (2018) 'Wild Patterns: Ten Years After the Rise of Adversarial Machine Learning.' Pattern Recognition 84 — the field's consolidated, sober consensus.
- NIST (2025) 'Adversarial Machine Learning: A Taxonomy and Terminology' (AI 100-2e2025) — the shared vocabulary for evasion, poisoning and their variants.
- Greshake, K. et al. (2023) 'Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection.' AISec '23 — the naming and demonstration of indirect injection.
- OWASP GenAI Security Project (2025) 'OWASP Top 10 for LLM Applications' — prompt injection and related integration risks as leading concerns.