ISO 27090 Threat Cards: Developer Edition
The 12 ISO27090 AI system threats plus 6 privacy and 6 resilience threats, written for developers, with vulnerable code to spot in review.
The standard threat cards describe what can go wrong. This edition shows what it looks like in code. Each card pairs one threat with a vulnerable pattern in Python 3.12 and TypeScript 5 on Node 22: the kind of code that passes a quick review and should never ship.
Every card names where the threat shows up (data loaders, scoring APIs, LLM gateways and so on), what an attacker needs, and the controls that fix it. It ends with a code review checklist and the tests that prove the fix works.
What's in the deck
- 12 AI security threats (T01 to T12) from ISO/IEC 27090 Table 1, from data poisoning to output that carries injection attacks
- 6 privacy threats (P01 to P06), including re-identification of de-identified data, erasure and rectification failure, and unlawful solely automated decisions
- 6 resilience threats (R01 to R06), including sponge inputs, cost exhaustion and LLMjacking, silent drift, and cascading failure in agent chains
The code samples are deliberately vulnerable. Use them to train reviewers and to seed test cases, not as a starting point for production code.
Fixes marked AIC-01 to AIC-16 are described on the ISO 27090 Control Cards. For a non-technical audience, use the original ISO 27090 Threat Cards.