Education Secretary Linda McMahon defended carefully guarded AI use after CNN asked whether children were “guinea pigs for chatbots.” Her answer echoed new federal guidance demanding proof of learning gains, but her praise for Alpha School went further than independent evidence currently supports.
.jpg){kind=link}
Education Secretary Linda McMahon walked into a fast-moving debate Sunday with two positions that can coexist only if schools take the evidence requirement seriously: artificial intelligence may become a useful classroom tool, and educators still do not have enough long-term evidence to know where its benefits end and its risks begin.
During CNN’s State of the Union on Aug. 23, host Dana Bash pressed McMahon on that uncertainty. McMahon did not dismiss the concern. She acknowledged that “there aren’t a lot of metrics available now,” while arguing that AI can be introduced slowly, with guardrails, monitoring and a willingness to remove products that fail to improve learning.
McMahon’s answer was more nuanced
The exchange was sharper than a routine technology interview because Bash framed the issue around children becoming experimental subjects for a rapidly changing industry. McMahon responded by pointing to Alpha School in Austin, Texas, where she said students spend the first part of the day on computers and receive personalized instruction while adults monitor their progress.
McMahon described the model as offering something close to a one-on-one tutor. If a student falls behind, she said, material can be repeated; if a student moves quickly, the program can accelerate. She called the technology a “great tool” at that school but also said schools should ask who uses AI, for how long, and with what outcome.
McMahon did not say AI should replace teachers. Later in the same interview, she said human interaction is “incredibly important” and that nothing should replace direct teacher-student interaction. She also endorsed removing recreational screen distractions while distinguishing them from screen time used for instruction.
The tension came from how confidently she described Alpha’s results while admitting the broader evidence base remains limited. McMahon called the model “incredibly effective” for those students. Yet a visit, student conversations and a school’s reported outcomes are not the same as independent evidence showing that a model will work across different ages and school systems.
Federal policy is pushing AI
The administration is not approaching classroom AI as a neutral observer. President Donald Trump’s April 2025 executive order made AI literacy and proficiency an explicit federal policy goal and called for “appropriate integration” of AI into education, teacher training, public-private partnerships and research aimed at improving student outcomes.
The Education Department moved further in April 2026 by establishing “Advancing Artificial Intelligence in Education” as one of the secretary’s supplemental priorities that can be used in discretionary grant competitions. AI is therefore an area the federal government has formally chosen to encourage within existing education programs.
But on Aug. 20, three days before McMahon’s CNN appearance, the department issued guidance that sounded more cautious than a technology sales pitch. It told states and districts to prioritize instructional value, preserve educator judgment, provide transparency to parents and demand evidence that technology is producing meaningful learning outcomes.
The guidance distilled that approach into five questions: What learning problem does a tool solve? When should it be used? For whom? For how long? And what evidence shows that it improves learning? It also said schools should change course when evidence is weak and remove tools when repeated findings show persistent shortcomings.
Alpha School is the test case
Alpha School has become a prominent example because its model is more ambitious than adding a chatbot to homework help. Alpha says students can complete core academic work in two hours a day through adaptive technology and mastery-based learning, leaving the rest of the day for projects, life skills, sports and other activities.
The school reports eye-catching results. Its website says students average 2.6 times the growth of similarly scoring peers on nationally normed MAP assessments and that a majority perform near the top of national distributions. Those are Alpha’s claims, based on data it presents from its own students; they are not an independent randomized evaluation of the model.
That distinction matters when a federal official presents Alpha as evidence that AI is working well. Its outcomes may reflect many variables beyond software, including admissions, family resources, student motivation, staffing, school culture and individualized adult attention surrounding the technology.
Even Alpha’s own model complicates the shorthand that students are simply being taught by chatbots. The school describes a broader system combining adaptive technology, mastery learning and human “guides.” The policy question is whether a technology-heavy instructional system can reliably produce durable learning without weakening the human parts of schooling.
Research cuts both ways
There is more evidence about AI-supported learning than the phrase “no real studies” might suggest, but it remains uneven. A 2025 systematic review of 28 studies involving 4,597 K-12 students found that AI-driven intelligent tutoring systems generally had positive effects on learning and performance, while advantages were smaller when compared with non-intelligent tutoring systems.
The authors called for longer studies, larger and more diverse samples, and greater attention to ethical questions. Much of the older “AI tutoring” literature also concerns structured intelligent tutoring systems, not the open-ended generative chatbots now being integrated into products used by students.
A Harvard randomized controlled trial published in 2025 found that college students using a carefully designed AI tutor learned more in less time than students in an active-learning physics class and reported higher engagement and motivation. But the experiment involved university students, a specific subject and a tutor deliberately built around established teaching practices.
Those results show that AI can support learning under some conditions. They do not establish that a general-purpose chatbot is effective for elementary students or that the same gains would survive when a tool is scaled across thousands of classrooms with different teachers, curricula and student needs.
Guardrails can change the result
One of the strongest warnings comes from a large high-school mathematics experiment led by researchers at the University of Pennsylvania. Nearly 1,000 students in Turkey were assigned to use no AI, a standard GPT-4 interface, or a specially designed GPT tutor during practice sessions.
During practice, access to GPT-4 helped students answer more problems correctly. The problem appeared when the AI disappeared. Students who had used the more open-ended version performed worse on a later unassisted test than students who never had AI access, while the specially guarded tutor substantially reduced that negative effect.
The mechanism is intuitive: a tool that supplies answers can improve immediate task performance without building the knowledge a student needs when the tool is gone. A tutor that prompts reasoning, limits shortcuts and is designed around instructional goals can behave differently. “AI in education” is too broad a category to have one universal effect.
That finding puts weight behind McMahon’s phrase “AI with guardrails.” The unresolved question is who defines those guardrails, who tests them, how quickly weak products are removed and whether schools have the expertise to demand evidence from vendors before a technology becomes embedded in everyday instruction.
Educators are not simply anti-AI
The criticism that followed McMahon’s interview was real, but it should not be stretched into a claim that educators uniformly reject artificial intelligence. Raw Story highlighted posts from teachers and advocates who argued that AI should not displace human teachers and criticized McMahon for praising a technology-heavy model while acknowledging the need for more evidence.
That skepticism fits a broader labor position, but the American Federation of Teachers is not calling for a blanket rejection of all AI. The union operates an AI training initiative and says technology can enhance teaching when educators remain central, student safety and privacy are protected, and schools use explicit guardrails.
At the same time, the AFT’s 2026 policy is far more restrictive for younger children. It advocates no screens for prekindergarten through second grade except for compelling needs, opposes student-facing AI in elementary schools, and calls for strict protections for older students. It also wants independent research rather than evidence financed only by technology companies.
The National Education Association has taken a similarly process-focused approach. Its guidance asks who chooses AI systems, how student data are protected, whether humans verify AI-generated content and grades, and how districts will reevaluate tools. It specifically asks how schools will ensure AI supports rather than supplants human-directed teaching.
Why the backlash landed
McMahon’s answer irritated critics because the burden of proof is unusually important when children are involved. Schools cannot treat a product like a consumer app that can be abandoned with little cost. Once districts train teachers, sign contracts, redesign lessons and normalize a platform, removing it can become financially and institutionally difficult.
The secretary’s own department recognizes that problem. Its guidance tells schools to demand evidence before allowing technological novelty to substitute for instructional value. The difficulty is applying that rule consistently while an administration is simultaneously promoting AI literacy, directing grant priorities toward AI and celebrating high-profile models not independently validated at public-school scale.
There is also a difference between teaching students about AI and placing AI between students and teachers for core instruction. The Trump executive order supports AI literacy, workforce preparation and teacher training. Those goals do not logically require schools to turn elementary lessons over to chatbots or adopt a two-hour software-centered model.
A school can teach students how AI works, use AI to help teachers plan lessons, deploy a tightly constrained tutor for a specific skill and still reject open-ended chatbot use for younger children. Those are separate choices with different risks.
The real test comes next
McMahon’s CNN interview did not settle whether students are becoming “guinea pigs.” It clarified the standard by which the administration says classroom technology should be judged: evidence of learning, limited and purposeful use, teacher involvement, parental transparency and removal when a tool fails. The challenge is whether policy decisions will actually follow that standard.
For school leaders, the available research supports carefully designed AI tutoring in some settings and warns that poorly designed systems can weaken independent learning. Age, subject, instructional design, data practices and the continuing role of human educators all change the risk-benefit calculation.
For McMahon, the political vulnerability is narrower. She has made Alpha School a positive example while conceding that the field lacks many of the metrics needed to judge newer AI products. If the administration wants schools to demand evidence from vendors, it will face the same demand when it points to particular schools as models.
That is why the controversy extends beyond one Sunday interview. The federal government is trying to accelerate AI literacy while telling schools not to confuse innovation with effectiveness. Whether those goals can coexist will depend on rigorous evidence, transparent evaluation and a willingness to stop when the technology does not help students learn.
Leave a Reply