CASE_FILE_05 //

OPERATION:
GATEKEEPER

REAL NAME: LLM Security Firewall

CLASSIFICATION: adversarial containment

STATUS: IN_DEVELOPMENT

PERIOD: 2025

ROLE: researcher · engineer

STACK: Python · OWASP LLM Top 10 · FastAPI · Redis

the problem

LLMs are deployed everywhere with no inspection layer between them and the user. Every input is a potential exploit; every output is a potential leak.

the approach

A reverse proxy that sits in front of the model. Inspects inbound prompts against the OWASP LLM Top 10. Inspects outbound completions for sensitive patterns. Logs everything. Blocks the obvious; flags the suspicious for review.

what was built

[in progress]

what was learned