Nine detectors in parallel. One verdict.
AEGIS is a purpose-built classification engine for content engineered to evade filters. Each detector specializes; the ensemble decides. Built for the adversarial environments where LLM-wrapper classifiers fail.
What AEGIS is
AEGIS is a content classification engine. It scores content on a 6-axis weighted rubric using dual-judge LLM dimension scoring (Gemini 3 Pro + Claude Haiku 4.5 as primary judges in the analyze pipeline), with a structured evidence trace attached to every verdict. Cross-modal correlations and additional purpose-built detectors are on the research roadmap.
The design assumption is adversarial. Content that arrives at AEGIS is content that may have been engineered to get past a filter. Encoded payloads, cross-modal tricks, context collapse, and signal inversion are the working environment, not the edge case.
How the rubric composes
A 6-axis weighted rubric scores each piece of content across visual, audio, text, metadata, semantic, and contextual dimensions. Two LLM judges (Gemini 3 Pro and Claude Haiku 4.5) evaluate each dimension independently; their scores are combined into a single verdict with a confidence band and a structured evidence trace. Cross-modal, linguistic, and behavioral detectors are framed as research / roadmap.
The rubric is auditable. Every verdict carries the per-dimension scores from each judge, so a downstream consumer (a parent dashboard, a partner SOC, a regulator) can inspect the basis for a hold, a flag, or a clear.
Why dual-judge rubric scoring beats single-shot moderation
Single-shot moderation calls an LLM once and returns a label. AEGIS uses two independent judges (Gemini 3 Pro and Claude Haiku 4.5) scoring each rubric dimension, then composes their outputs into one verdict with attached evidence. The pipeline is updateable in response to specific evasion patterns by re-weighting the rubric and refreshing the judges, without retraining a foundation model.
Practically, this means AEGIS responds to new evasion classes in days; produces evidence consumers can audit; and offers a v1.1 roadmap that adds purpose-built detectors and cross-modal correlations alongside the rubric judges.
What AEGIS replaces
AEGIS replaces the legacy DNS-blocklist layer that residential ISPs, schools, and family-safety apps rely on. A blocklist is a list of bad domains; AEGIS classifies content. A blocklist is a binary verdict; AEGIS attaches evidence and confidence. A blocklist is updated on a curator's schedule; AEGIS updates against patterns the engine itself surfaces.
It does not replace human review. It replaces the rituals of pretending a blocklist is enough.
Where it runs
Four deployment modes, four licensing paths. Edge in ISP infrastructure. Embedded in OEM hardware. Embedded in retail routers. Hosted as a service for enterprise content platforms.