Research and methodology
Methodology, published. Weights, retained.
Public methodology papers, ongoing benchmarking disclosures, and research-license access to corpora for academic partners. The exact detector weights and corpus contents stay proprietary.
Whitepaper · 2026
AEGIS architecture overview
An overview of the AEGIS classification engine: nine purpose-built detectors running in parallel, an ensemble verdict layer, and the deployment topology that lets the same engine ship as a network-edg...
Read →
Methodology paper · 2026
Adversarial content classification: lessons from operating at web scale
What a decade of operating content-safety systems on a 45M+ monthly-active-user surface teaches about adversarial classification. Covers signal inversion, encoding tricks, cross-modal evasion, context...
Read →
Technical paper · 2026
Cross-modal evasion and how to defeat it
A taxonomy of cross-modal evasion patterns: cases where each individual modality reads benign but the combination is unsafe. Describes how AEGIS's cross-modal detector composes per-modality signal int...
Read →
Ongoing disclosure · 2026
Benchmarking AEGIS against public content-safety datasets
An ongoing disclosure of AEGIS performance against publicly available content-safety datasets. Reports structural results (which datasets, which detector configurations) without publishing specific pr...
Read →