KAVACH — MIRROR

Every message arrives in a language.
Mirror reads what's underneath.

Two passes. The first reads language — Mirror checks the same story across every major Indian script, and English, and traces where each one came from. The second reads pixels — for anything audiovisual, it finds the seams an AI leaves behind.

STEP 01 CROSS-LINGUAL PROVENANCE

Same story. Every script. One truth.

The fastest way to detect a fabricated message is to compare it with the same story, told elsewhere. Mirror pulls the matching headline from verified sources in every major Indian script — and English — then checks each one's origin and routing. The imposter is the one whose paper trail doesn't match.

Most verify.
A FEW DON'T ·
MEDIAN SESSION
● STEP 01 · PROVENANCE SCAN MIRROR / PIPELINE-A
MIRROR / TRACE 15 MAY 2026
STEP 02 GENERATIVE FORENSICS

For anything that moves, Mirror looks closer.

Once a piece of audio or video is flagged, Mirror performs pixel-level and phoneme-level forensics. It greets the content in its own languages, then runs a generative-artifact sweep — seam-finding, latent-noise detection, voice-formant trace. A synthetic face fails in ways a real one doesn't. That's where the mask slips.

High-confidence.
INDIAN-LANGUAGE
DEEPFAKES
● STEP 02 · GENERATIVE FORENSICS MIRROR / PIPELINE-B
MIRROR / TRACE 15 MAY 2026
Every major script.
Devanagari · Tamil · Bengali · Telugu · Gujarati · Kannada · Marathi · Punjabi · English — and expanding.
— INDIC + LATIN
Sub-second
From ingest to verdict on short clips
— TYPICAL
Human-validated
Every flagged item reviewed by a language-native analyst before release
— SOP