Version History
Changelog
A chronological record of what has shipped. Aiscern is actively maintained — this page updates with every release.
v4.1.0
April 2026
LatestUpdated image detection model and audio accuracy improvements.
- ImprovedImproved audio detection accuracy to ~79% on WaveFake benchmark
- ImprovedUpdated image ViT ensemble with additional GAN artifact training
- NewPDF report export for Pro users (beta)
- FixFixed video detection timeout for clips longer than 2 minutes
- FixFixed broken video fallback that was suppressing AI verdicts
- InfraMigrated rate limiter to persistent storage — resets correctly across cold starts
v4.0.0
March 2026
Major ensemble rewrite. Detection signal logic overhauled across all modalities.
- NewComplete rewrite of detection signal logic in hf-analyze.ts (v5.0.0)
- ImprovedMulti-model consensus now weights by modality — no single model dominates verdict
- ImprovedText detection ensemble updated with DeBERTa-v3-base
- NewAudio detection added (ElevenLabs, common TTS tools, voice clones)
- FixFixed image buffer truncation that was corrupting files sent to HuggingFace
- FixFixed duplicate database writes on concurrent scans
- FixFixed in-memory rate limiter resetting on Vercel cold starts
v3.2.0
February 2026
Batch processing, API access, and scan history.
- NewBatch scan — process up to 20 files simultaneously
- NewREST API v1 endpoint for text detection (Team+)
- NewScan history saved for signed-in users
- ImprovedConfidence threshold tuned: ≥62% AI, ≤38% Human
- FixHero card CSS opacity bug fixed on mobile Safari
- FixAPI key management — fake validation removed, real key auth implemented
v3.0.0
January 2026
Video detection added (experimental). ARIA detection assistant launched.
- NewVideo deepfake detection — experimental, ~76% accuracy
- NewARIA — AI Detection Assistant chatbot
- NewScan share links — shareable results with one click
- ImprovedImage detection improved with CLIP-based signal analysis
v2.0.0
December 2025
Initial public launch. Text and image detection live.
- NewText detection (RoBERTa ensemble) — ~85% accuracy
- NewImage detection (ViT + GAN artifacts) — ~82% accuracy
- NewFree tier live — 10 scans/day, no credit card required
- NewMethodology page published with accuracy benchmarks
Found a bug or want to suggest a feature? Contact us or check the roadmap.