Antisemitism and bias detection API for DSA compliance

Scan articles, comments, and transcripts for antisemitism, anti-Israel bias, and context omissions. IHRA-grounded, attribution-aware, with full audit trails.

Get started free Read the docs

DSA Article 34 ready

| IHRA Working Definition | Three-track analysis | Attribution-aware scoring

The compliance gap in content moderation

The EU Digital Services Act created obligations. It did not create the tooling to meet them.

of global revenue

Maximum DSA fine for non-compliance with systemic risk obligations.

established standards

The EU has published no official technical standard for bias compliance — yet.

87%

self-reported

Current DSA transparency reports rely on platform self-certification — the system regulators are trying to replace.

How it works

One REST endpoint. JSON in, structured analysis out. Every detection is persisted for audit.

Ingest

Send any text — an article, a comment, a transcript — to POST /v1/detect with your API key. Optional title and source fields sharpen the analysis.

Detect

A four-stage pipeline runs: dog-whistle scan, LLM triage, three-track deep analysis (antisemitism, anti-Israel bias, context omissions), then adversarial verification. IHRA-grounded, attribution-aware.

Audit

Every detection is persisted with confidence scores, IHRA references, character-level snippets, and a routing decision: auto-accept, human review, or auto-reject.

Terminal

curl -X POST https://magendetect.com/v1/detect \
  -H "x-api-key: mk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"title": "...", "source": "...", "content": "..."}'

# → {
#     "status": "flagged",
#     "detections": [{
#       "category": "antisemitism",
#       "subcategory": "classic_trope_conspiracy_control",
#       "confidence": 0.95,
#       "attribution": "publisher_voice",
#       "definition": { "framework": "IHRA", "reference": "IHRA-7" }
#     }],
#     "calibration": { "review_decision": "auto_reject" }
#   }