Amid the AI Panic, Which AI Tools Should We Let Access Student Data?
AI policies for schools this year faced emerging issues with data privacy.
In February 2026, federal agents raided the home and office of LAUSD superintendent Alberto Carvalho. The investigation wasn't financial, but rather about a chatbot shaped like a smiling sun.
Three years earlier, Los Angeles Unified had signed a $6 million contract with a pioneering ed-tech startup called AllHere to build an AI chatbot. It was meant to surface grades, attendance, and test scores to families.
Then it came apart.
By mid-2024, the startup had filed for bankruptcy and its CEO ran off. Chris Whiteley, a former AllHere software engineer, also claimed that student data was sent overseas, over-collected, and logged by third parties.
"Senior executives didn't understand how to protect data," said Whiteley.
LAUSD's AI disaster isn't an outlier so much as a precedent.
While districts across the country race to establish their AI policy, smaller details and problems continue to emerge: actually implementing AI policies, reducing shadow AI, and managing grey situations all contribute to the current AI panic and crackdown.
The question for us boils down to: What should we do before the next “AllHere” happens to us?
Researchers say focus on today's harms, not the AI apocalypse
In September 2026, Anthropic's Dario Amodei publicly argued for pacing AI's capability gains so safety can keep up, implying that AI progress is moving faster than what safety protocols can keep up.
But the strongest consensus among researchers, reported in Education Week's examination of the AI panic, is that this doomsday framing distracts from real, present problems.
Justin Reich (MIT Teaching Systems Lab) argues educators already have enough to address in the harms happening now (data privacy, cognitive offloading, deepfakes) without speculating about the apocalypse. He's also openly skeptical of the tech industry's motives in amplifying existential fear.
Rebecca Winthrop (Brookings Institution) reframes the slowdown as an opening to do what she calls narrow but genuinely useful AI: tools that flag in real time where students are struggling. Her sharpest guidance? Steer clear of AI agents in education, the proactive, multi-step "agentic" kind, because it hasn't been tested in schools.
Jeremy Roschelle (Digital Promise) predicts the fear will push more districts toward bans, but also sees a chance to stop buying hundreds of unvetted tools he memorably calls "vibe coded," built fast with too little oversight.
AI management is maturing fast, but with real gaps
79% of districts now have AI guidelines, up from 57% a year earlier, per CoSN's 2026 State of EdTech report (607 district technology leaders). The share with no guidance fell from 43% to 21%.
Cybersecurity remains the #1 priority, as it has for eight straight years.
37% of districts plan to strengthen vetting of third-party agreements involving student data.
But maturity is uneven.
By one analysis of the same data, only about 34% of those AI policies actually address data privacy. The policy document exists; the depth underneath often doesn't. A guideline that never says where student data goes is incomplete.
Vetting AI tools is now the IT director's job
After the AllHere incident, Leslie Eaves of the Southern Regional Education Board put it plainly: "Anything with AI is a risky click."
Her commission's advice? Let adults test tools thoroughly before students touch them (which reflects a field that got blunt about risk if we're being honest).
The problem is scale.
Reporting on AI procurement describes technology officials swamped by the sheer volume of tools and the speed of vendor updates. New federal guidance is pushing districts to show evidence of instructional impact before they buy or renew.
Helpfully, shared infrastructure is emerging:
EdTech Quality Indicators Guide: five pillars, the first being "Safe": data minimization and transparency.
CoSN's K12 Community Vendor Assessment Tool: a standardized risk questionnaire built for RFPs.
State guidance: Pennsylvania, for example, tells schools to know who controls the data, limit collection, and keep a human in the loop for grading or discipline.
The working rule:
If a tool touches a student record, it's a procurement decision that needs a written agreement before a pilot, not after.
For IT leaders who want a structured way to start these conversations with their teams, use Deledao's free AI-readiness card deck for technology directors covering infrastructure, compliance, and vendor vetting.
The gap between having an AI policy, and implemented protection
There is no AI exception to FERPA: The moment a tool processes a student education record, the law applies.
Generative AI adds new risks: vendors may use student inputs to train commercial models, tools may retain inputs indefinitely, and a teacher signing students up for a free consumer chatbot can create an unauthorized disclosure.
The FTC's 2025 COPPA amendments: now require separate parental consent to share children's data with third parties.
The exposure is human: ~60% of teachers used AI in 2025, but about two-thirds got no training (which is how "shadow AI" takes hold)
The defensible posture privacy specialists keep converging on: get written Data Privacy Agreements, demand compliance documentation, and treat vendor reluctance as a red flag.
It's worth noting this is exactly the posture the AllHere case punished the absence of: offshore storage, over-collection, third-party logging.
Fortunately, some proactive vendors have built their whole system against avoiding these privacy failures.
Deledao, a K12 web filter built on data privacy, keeps student data in U.S. data centers and exceeds compliance requirements (including ISO/IEC 27001 certification). It acts as a data processor rather than an owner and never sells PII.
What teachers and students actually think about AI in schools
The sentiment data complicates the tidy "ban it or buy it" binary.
Teachers struggle to know if students are learning or using AI
A report from Harvard's Center for Digital Thriving (surveys of 1,000+ teachers and principals, plus teen interviews) found:
74% of teachers and 69% of principals named a cheating-related issue as their main AI dilemma.
Fewer than 20% of teachers work at a school with a specific generative-AI policy.
The fixation on AI cheating in schools is eroding trust. Co-founder Beck Tench's line should resonate with anyone who thinks about classroom visibility: "Teachers can't see their students anymore."
Policies framed around suspicion, the researchers warn, just push use underground into what Wharton's Ethan Mollick calls the "secret cyborg.”
A separate survey of 303 Wisconsin educators (including IT and technology directors) found the deeper worries were increased student reliance on AI and reduced critical thinking: the evidence-of-learning problem.
Students report feeling pressure to hide their use
An estimated 84% of high schoolers used generative AI for schoolwork in 2025.
Students report feeling pressure to hide their use, but also describe real value in chatbot tutors.
Districts like Oregon's Umatilla are bringing students into policymaking; its students helped draft a national AI framework.
Meanwhile, public wariness is rising, with parent pushback and even visible anti-AI protests.
The AI policy whiplash of bans, pauses, and opt-outs
At least eight states now require districts to have an AI policy. The landscape is a patchwork, which is itself the lesson:
Los Angeles Unified restricted AI use for all students.
New York City imposed a moratorium through 8th grade, with twice-yearly AI literacy for high schoolers.
Florida called for districts to let families opt in.
No single model is worth copying wholesale. A defensible local posture beats waiting for a consensus that isn't coming soon.
What it means for K12 IT leaders. How should we move?
Favor narrow, proven tools over flashy ones
Prioritize specialized deterministic AI tools that exceed compliance and regulation. Avoid broad, generative, and abstract tools. Also vet like your signature is on the disclosure, because functionally it is. DPAs, compliance docs, human-in-the-loop, evidence of impact. You can't support what you can't see, but visibility and surveillance aren't the same thing.
Create open feedback loops with teachers (and students)
Shutting down shadow AI only eliminates part of the problem. Those tools came into use because it solved teacher problems. Therefore, see what AI tools they want to use, and consider vetting those tools into your school. If not adequate, offer a better alternative, so no one's driven to a risky free tool again.
The most grounded voice may be Superintendent Umatilla Heidi Sipe. She frames the job not as predicting the future or chasing hype, but as developing strong learners who can think critically, and trusting students enough to help decide AI's role.
For us technology leaders, that's our assignment under all the apocalyptic headlines: be the one in the building who actually knows what's inside the tools, so that good, careful use is even possible.
