GitHub Pages edition. Scoring, character attacks, normalization, and bounded email preview run entirely in this tab. Run the full local API for RFC-complete triage, semantic attacks, and strict LureProof evidence.

Defensive AI engineering / live workbench

Break the detector before they do.

Score a suspicious message, mutate it like an attacker, then verify whether your defense earns the catch back.

01 / Input

Message under review

Processed locally

Message text

Detection engine

02 / Readout

Detection signal

Awaiting analysis

Run the message through a detector
to reveal its fraud signal.

—%

BENIGN / 0THRESHOLD / 50FRAUD / 100

Detector—

Threshold source—

Contributing terms

Real-world workflow / RFC 5322

Triage a suspicious email file

Choose a saved .eml. LureScope reads its message text, headers, link names, and attachment names locally. It never visits links or opens attachment contents.

Choose .eml file

—

——

—

No message content or addresses are written to the proof.

03 / Stress test

Run the adversarial sequence

No external calls for character attacks

Attack technique

Countermeasure

01 / Original—

02 / Attacked—

03 / Defended—

04 / Evidence Explorer

Open the proof, not the message

Inspect a LureEval receipt, Pilot Gate, Defender comparison, Shadow report, LureProof, or SCuBA statement. Parsing stays in this tab. The explorer reveals metrics, privacy boundaries, byte bindings, and limitations without uploading the file.

Choose evidence JSON

—