Implement output validation with Pydantic/JSON Schema

1. 01

   Concepts & system design

   60 min· read · map the mechanismtimer

   R07 — JSON Schema strict validation

   R30 — OWASP LLM05: Improper Output Handling

   Read the assigned sections; map the mechanism, threat model and trade-offs.

2. 02

   Project implementation

   120 min· build · integrate · debugtimer

   Build output-contract (D6): strict server-side Pydantic/JSON Schema validation, enum/range/length constraints, unexpected-field rejection (extra='forbid'), and sink-specific HTML/Markdown sanitization.

   Build, integrate, debug and improve the increment in the active project.

3. 03

   Security lab

   60 min· attack your own systemtimer

   Inject XSS payloads (<script>, javascript: links, malicious Markdown image beacons) into simulated model outputs and prove D6 neutralizes them before browser/API rendering.

   Test the abuse case, failure path, authorization bypass or resource limit.

4. 04

   Verification & evidence

   30 min· test · record evidencetimer

   D6 output validation and sink-sanitization test suite.

   Add or run regression tests; record the 9-point evidence record.

5. 05

   Deliberate practice & Socratic review

   30 min· explain aloud · update gapstimer

   Answer the 8 Socratic questions without notes: What? Why? How? What if? Why not? How can I prove it? Where can it break? What have I learned?

   Re-explain the hardest concept aloud, inspect one edge case, update the gap log.

6. total5 h

• R07JSON Schema↗ json-schema.org

  Core schema types, required properties, additionalProperties: false, enums, and numeric/string bounds.

• R30OWASP LLM Output Handling↗ genai.owasp.org

  Improper output handling (LLM05): pairing schema validation with sink-specific encoding (XSS/SQLi/SSRF prevention).

Read only the named sections needed for today's task. Reading sits inside the 60-minute concepts block; it is not extra time.

• docsJSON Schema (official)
  Getting Started Step-By-Step

  Builds a schema incrementally with required properties, type constraints, and validation runs — the contract you mirror in the D6 output schema.

  R07checked 2026-11-11
• docsPydantic docs
  Strict Mode

  Explains the coercion that silently turns '123' into 123 and how strict mode stops it, which is required so model output cannot smuggle in wrong-typed fields.

  R07checked 2026-11-11
• videoLearn with Amine
  How do you use Pydantic v2 for validation and structured outputs with LLMs

  Shows Pydantic v2 models used as the LLM output contract with validation errors surfaced, i.e. schema-first parsing rather than regex on prose.

  R07checked 2026-11-11
• videoAnkit Bansal
  Data Validation for Data & AI Engineering Using Pydantic | Pydantic Basics to Structured Output

  Covers field constraints, validators, and using Pydantic to get reliable structured output from models, the enum/range/length enforcement in D6.

  R07checked 2026-11-11
• articleOWASP Gen AI Security Project
  OWASP LLM05: 2025 Improper Output Handling

  Explains why schema validation alone is not enough and pairs it with sink-specific encoding to stop XSS, SQLi, and SSRF through model output.

  R30checked 2026-11-11

A lab is complete only when it produces evidence. Installing PyRIT, Garak, Presidio or TransformerLens is not completion. Saved on this device only.