New Mexico's AI Reading Screener Cuts False Dyslexia Flags, Faces Backlash
An AI tool that listens to kids read aloud sharply reduced false dyslexia flags in New Mexico, but parent pushback over privacy is putting it on the chopping block.
What happened: New Mexico's education department has been defending its use of Amira Learning, an AI-powered reading screener, after parent groups in Santa Fe and Albuquerque organized against it, citing privacy worries and general distrust of anything labeled "AI." Adopted in 2025 under a 2018 court order to improve outcomes for underserved students, the tool listens to K-2 children read aloud and flags hesitations, substitutions and decoding errors. Every district can now opt out. The fight escalated in September when New York City paused generative AI for about 600,000 students and Los Angeles pulled it from every district device - decisions aimed at chatbots that critics are already invoking against non-generative tools like Amira.
Why it matters: New Mexico has the lowest fourth-grade reading scores in the country, and research cited by the Annie E. Casey Foundation finds children not reading proficiently by third grade are four times more likely to eventually drop out of high school. State data shows the screener is fixing a real accuracy problem: last year it flagged roughly one in six first graders for possible dyslexia, matching national rates, versus 45 percent under the prior computer-based test. That means nearly half of six-year-olds, often non-native English speakers, had previously been mislabeled - a gap that decides who actually gets extra reading help.
What critics are right to ask: Parents and 35 state lawmakers have pressed for answers on what happens to children's voice recordings, and Los Alamos and Santa Fe paused the tool over concerns recordings could train Amira's commercial products. Speech-recognition AI is also documented to perform worse on accented or atypical speech, raising fears that English-language learners could again be mis-sorted through a new mechanism. These are legitimate questions about consent, retention and bias testing - but advocates argue they're reasons to demand fixes and proof, not to scrap a tool that has already measurably cut error rates.
The rollout: Albuquerque offers one model: it kept the screener, explained to parents why recordings are collected, and shortened how long data is stored. Other districts have paused the tool or opted out entirely under the state's new flexibility. With New York City and Los Angeles's generative-AI bans now cited in other state capitals, more districts are expected to face the same choice this year - banning anything called "AI" outright, or writing specific rules on data use, accuracy and bias for each tool individually.