AI Efficiency in Schools Conceals Privacy Failures
The promise arrived with clean interfaces and teacher testimonials: AI would save time, personalize learning, close gaps. Classroom adoption jumped fast. In 2025, RAND’s national survey found 54% of students and 53% of teachers using AI for school — both up more than 15 percentage points from the year before Forbes article. The tools were deployed because they worked, or at least seemed to. But the deception lives in what was left unspoken: who gets to see the data, what happens to it once the tool runs, and who’s accountable when the vendor changes hands or suffers a breach.
The Moment Efficiency Became the Opposite of Progress
In December 2024, PowerSchool reported a data incident that compromised student and teacher records. The exact number of affected records was not disclosed by the company, but the breach exposed sensitive information including names, addresses, and disability statuses Forbes article. The system was built to streamline grades, attendance, and enrollment. It made administrators’ lives easier. But the same centralized data that enabled efficiency also created a single point of failure. The promise of progress turned into the opposite: a catastrophe that left families guessing whether their child’s name, address, disability status, or behavioral records were now in the hands of criminals.
Spring 2026 brought another incident. Canvas, the learning management platform used by thousands of schools, suffered a breach that exposed a variety of personal information from many institutions Forbes article. Again, the tool was sold as a convenience: post assignments, submit work, communicate with teachers. The efficiency was real. The protection was not. These were not small districts with limited resources; these were platforms adopted by entire states and large urban districts.
A key deception is that schools often adopted these tools because they believed the vendor’s assurances. Sales materials emphasized data encryption and compliance with vague standards. But encryption is irrelevant if authentication fails, and compliance becomes meaningless when the standard itself was written before the internet existed. The real question is not whether a vendor promises privacy, but whether the school has actually verified those promises through signed agreements and independent audits.
The Feedback Loop Between Deployment and Deception
Policy did not keep pace with usage. In 2026, only 45% of principals reported having a school or district AI policy, and just 34% of teachers said their district had a policy addressing AI and academic integrity Forbes article. The gap between adoption and governance creates a dangerous feedback loop: the more tools are used, the more data accumulates, the harder it becomes to retroactively protect it. Each new breach pushes districts to write policies, but those policies often focus on acceptable use — what students and teachers can do — rather than on data handling, retention, or vendor oversight.
Ohio became the first state to require every public school district to adopt a formal AI policy by July 1, 2026, under House Bill 96 Forbes article. The state’s model policy does name data privacy directly, instructing districts to address protection of personally identifiable information and compliance with FERPA. But not every district uses the state model. Some write their own, often by copying a neighboring district’s document or hiring a consultant who recycles boilerplate language. The policy becomes a checkbox, not a shield.
Meanwhile, the federal law that supposedly protects student data — FERPA, the Family Educational Rights and Privacy Act — was written in 1974 Forbes article. It has no explicit cybersecurity requirements. It was never designed for AI models that train on the data they touch. A 50-year-old law cannot govern tools that did not exist when it was written. Yet many districts still point to FERPA as their primary safeguard, creating the impression of protection where none exists.
California’s AB 1159, still moving through the state legislature in mid-2026, aims to prohibit schools and educational technology vendors from using student data to train AI models Forbes article. The bill would also grant students and parents a limited right to sue if violations occur. But it faces opposition, and even if passed, it applies only to California. Idaho’s SB 1227 takes a narrower approach, requiring data privacy protections specifically for AI tools used in schools Forbes article. These bills exist because the current baseline does not cover this. But they are patchwork. A family in Nebraska or Florida gets no protection from California’s legislation.
The structural deception is that each new policy gives the illusion of progress while the underlying problem grows. Districts deploy AI tools, then write policies, then discover the tools do things the policies did not anticipate. The vendor changes its terms of service, or sub-processors appear, and the policy becomes a dead document. The feedback loop is not solving the problem; it is widening the gap between what schools claim and what schools actually control.
Whistleblowers and the Unseen Pipeline
In 2025, a whistleblower alleged that AllHere, a chatbot vendor used by school districts, improperly collected student data in violation of both industry standards and the district’s own written policies Forbes article. The chatbot was marketed as a friendly assistant that could answer questions about homework, schedules, and school events. Behind the interface, it was collecting far more than it needed. The district had a policy stating that data would only be used for the intended service. The vendor’s actual practices said otherwise.
These cases rarely make headlines unless a breach is massive or a whistleblower comes forward. Most violations go undetected because schools do not audit their vendors. The data privacy agreement — the signed contract between the district and the AI company — is supposed to restrict what the vendor can do. But compliance specialists say these agreements are often generic, lacking specific prohibitions on data reuse, retention limits, or requirements for breach notification Forbes article. A vendor can say “we value privacy” and then pass student data to a third-party model trainer, as long as the contract does not explicitly forbid it.
The deception runs deep: parents are told the school uses a “secure” AI tool, but security is a process, not a label. The tool may encrypt data in transit, but once the vendor’s sub-processor receives it, that encryption is irrelevant. The tool may claim to delete data after a student graduates, but deletion is only as reliable as the vendor’s internal procedures. And those procedures are almost never tested by an independent third party.
The Last Open Variable: What Happens to the Data After the Tool Stops
The most critical question remains unanswered for the vast majority of families: what happens to a child’s data when the child leaves the school or when the tool is discontinued? Many vendors keep data indefinitely, arguing they need it to improve their models or maintain service continuity. The school, having already signed the contract, has little leverage to demand deletion after the fact.
One task from the Forbes article asks parents to ask the school: “Ask how long the data is kept and whether it’s deleted when your child leaves the school or the tool is discontinued” Forbes article. This question is almost never asked at enrollment or at the time of tool adoption. It is the last open variable that determines whether a data breach is a temporary inconvenience or a permanent exposure of personal information.
The same burden falls on school leaders. They must go beyond writing a policy and actually verify compliance. The Forbes article advises parents to ask “who audits the vendor and how frequently this occurs” Forbes article. Most schools cannot answer that question. The ones that can are rare, and their answers often reveal that audits happen annually at best, or never at the vendor’s sub-processor level.
The deception is that efficiency — faster grading, personalized tutoring, automated attendance — has been prioritized over the infrastructure of trust. Schools race to write AI policies because they feel pressure to appear modern and responsive. But a policy that exists only on paper does not protect a single student’s data. It creates the appearance of safety without the substance.
The adoption curve is steep and not slowing: the same RAND survey that showed 54% of students and 53% of teachers using AI in 2025 also found usage rates climbing more than 15 percentage points year over year Forbes article. The deception is that the tools were brought in to help, and they do help — until they don’t. The breach at PowerSchool, the breach at Canvas, the allegations against AllHere — these are not anomalies. They are the predictable outcome of a system that values deployment over due diligence.
The last open variable is not whether schools will have AI policies. They will. The variable is whether those policies will be backed by enforceable contracts, independent audits, and meaningful consequences for failure. Without that, the efficiency of AI will continue to be the opposite of progress, and the students who trusted these tools will pay the price with their privacy.
Sources
1. PowerSchool
2. Canvas
3. California
4. Forbes