This month, first and second graders in San Francisco's public schools are taking California's required reading screening the way many children now do: by reading aloud to a computer. The software, Amira, listens, scores how the child reads, and flags those at risk of reading difficulties, including dyslexia. Similar tools are spreading because they promise something overstretched schools want badly, fast screening without pulling a teacher away for one-on-one testing. But in San Francisco and New Mexico, parents and teachers have pushed back over accuracy, privacy, and consent. If your child's school uses an AI screener, you don't have to choose between blind trust and opting out. Here's what these tools do, what the concerns are, and five questions worth asking.

What is an AI reading screener?

Amira, one of the most widely used, works by listening to students read aloud and catching their mistakes, according to an Albuquerque Journal report republished by Education Week in August. It also tutors, runs reading tests, and screens for dyslexia risk. New Mexico requires it for students in kindergarten through second grade, and the report lists it as authorized for state use in California, Georgia, Massachusetts, Michigan, Oklahoma, and Texas.

California's 2023 law requires schools to screen every child in kindergarten through second grade for risk of reading difficulties, including dyslexia, which we covered in our piece on California's screening law. A state panel approved a short list of screeners in December 2024, and Amira's maker says it received the panel's top endorsement. San Francisco Unified licensed Amira for every K–5 student in 2024, about 23,000 subscriptions for nearly $400,000 a year, The San Francisco Standard reported.

What are parents and teachers worried about?

The concerns fall into three groups, according to reporting by The San Francisco Standard and the Albuquerque Journal:

  • Recordings and consent. The software records children's voices. The Standard found San Francisco's contract never mentions recordings or how long they're kept, and the family notice for last year's rollout didn't mention AI or recording. Amira's chief executive told the Standard the district chose to delete audio monthly and that the company never sells student data. A Los Alamos, New Mexico, superintendent put the worry plainly in a public letter: "A child's voice is effectively biometric data."
  • Accuracy for some children. San Francisco teachers told the Standard the software struggled to understand students with speech issues and students with language-related IEPs, and some children reportedly had to "yell" to be understood. For a screener, a child the software mishears can look like a child at risk.
  • Evidence. Two research clearinghouses rate Amira's evidence as "moderate," the Standard reported, and the only randomized trial they reviewed tested a prototype, not the current product. A company-funded Louisiana study found small positive effects and low use.

There's another side. Albuquerque, which kept Amira, reported its highest first- and second-grade reading proficiency since at least 2021 last spring. That's encouraging, though it doesn't show the software caused the gains.

How have schools and states responded?

In New Mexico, Santa Fe and Los Alamos paused Amira in August, and Los Alamos switched to DIBELS, a screener a teacher administers in person. The state education department then issued a memo letting districts use Amira without voice recording, use a paper test, or use their own assessment for this school year, and setting a default of monthly deletion for recordings, the Albuquerque Journal reported. At least five districts and one charter school had declined the program by mid-August.

San Francisco Unified told the Standard it values student privacy and that families can opt their children out of the screener. A draft districtwide technology policy is scheduled for a first school board reading on February 9, 2027, with a vote planned for March 9.

What should parents ask about an AI reading screener?

  1. Which screener does my child take, and how is it scored? Ask for the tool's name and whether an AI or a trained adult scores your child's reading.
  2. Has it been checked for children like mine? If your child is learning English, has a speech or language need, or has an IEP, ask whether a person reviews low or unusual scores before they're used.
  3. What happens to the recordings? Ask whether audio is stored, for how long, who can access it, and whether it's used to improve the company's software. Ask to see the data agreement.
  4. Can I opt out, and what replaces it? Opting out should mean a different screener, such as a teacher-administered DIBELS or Acadience assessment, not no screening at all.
  5. What happens if my child is flagged? Ask what follow-up assessment comes next, what intervention starts, and how progress will be monitored.

What mistakes should parents avoid?

  • Opting out with no alternative. Early screening is one of the strongest protections a child with dyslexia has. Make sure your child is still screened another way.
  • Treating any screener score as a diagnosis. Screeners, AI or not, identify risk. Only a comprehensive evaluation identifies dyslexia.
  • Dismissing a flag because "the app is buggy." If the software flags your child, ask for a person to recheck with a different measure before deciding it was an error.
  • Assuming a high score settles it. If what you see at home matches the signs of dyslexia, request an evaluation in writing anyway; our advocacy toolkit has the letter.

We've long argued that screens serve children with dyslexia best as access tools, like audiobooks and text-to-speech, rather than as reading teachers, a view laid out in our guide to AI assistive tech. Screening is a different job, and the question isn't whether AI belongs in it. It's whether every child gets an accurate result, a person who looks at it, and help that follows.

Frequently asked questions

Do AI reading screeners record my child's voice? Many do. Tools like Amira listen to children read aloud and may store the audio. Retention varies by district contract; New Mexico's default is monthly deletion, with options for shorter or longer periods. Ask your school whether recordings are kept, for how long, and whether they're used to improve the software.

Can I opt my child out of an AI reading screener? Often, yes. San Francisco Unified says families can opt out of its Amira screener, and New Mexico lets districts use other assessments this year. If you opt out, ask in writing which alternative screener your child will take, so your child is still screened for reading risk, including dyslexia.

Is an AI screener accurate for children with speech or language differences? The evidence is limited. San Francisco teachers reported that Amira struggled to understand some students with speech issues or language-related IEPs. If your child has these needs, ask the school whether a trained adult reviews low scores and whether a teacher-administered screener can confirm the result.

Source: "Why kids, parents, and teachers are so frustrated by SFUSD's AI reading tutor" (Ezra Wallach, The San Francisco Standard, Sept 25, 2026) · "School Districts Push Back Against State-Required AI Reading Assessments" (Natalie Robbins, Albuquerque Journal via Education Week, Aug 4, 2026) · "New Mexico Allows Schools to Opt Out of Controversial AI Tool" (Natalie Robbins, Albuquerque Journal via GovTech, Aug 11, 2026) · Amira Learning announcement on California K–2 screening approval (company release, Dec 2024)