LatticeEthical AI Framework Explorer

Consumer platform · EU, US-Federal, BR · Owner: Jordan Vega

Aurora Content Moderation Pipeline

Multi-stage classifier and LLM judge for moderating user-generated content across hate speech, harassment, and CSAM categories.

System characterization

Lifecycle stage: monitoring
Autonomy level: autonomous
Security exposure: critical
Indigenous data: No
Affected populations: End users posting content, End users viewing content, Trust & safety reviewers
Created: 2025-09-04

Open framework items

Evade ML model
Active adversarial population; daily new evasion patterns observed.
High
LLM jailbreak
LLM-judge has been jailbroken in red-team exercises; monitoring tightened.
High
Transparency and explainability
Transparency reports published; appeals data not yet disaggregated.
Moderate
Recourse and redress
Appeals path exists; SLA review in progress.
Moderate
Manage
Active risk responses; continuous monitoring resourced.
Moderate