If you teach in a Google Classroom district, you have probably been told to check the compliance on any tool that touches student work before you use it. Fair. Student data in K-12 sits under FERPA, COPPA, and a stack of state privacy laws, and an AI-assistive grading tool reads student submissions to do its job. So security is the first question here, before anyone gets to whether the grading is any good.
Most AI grading tools describe their security in language that sounds reassuring and proves nothing. "Bank-level encryption." "We take privacy seriously." "FERPA friendly." A district IT reviewer cannot audit a vibe.
What a district can audit is independent verification: a named auditor, a Google review, a federal pledge with a public signatory list. I built ClassLens while teaching full time, and I decided early to compete on exactly this ground. We hold three independent security verifications, plus one public commitment you can check yourself. What each one actually means is below, along with where the line between them falls, and then a short checklist you can run against any vendor, us included.
Why compliance comes first with an AI grading tool
An AI-assistive grading tool does something most classroom software does not: it reads the actual content of student submissions to evaluate them. That is more sensitive than a tool that only stores names and grades.
FERPA lets a district designate a vendor as a "school official," which carries direct obligations about how that data is used and disclosed. If you teach students under 13, COPPA is part of the picture too, including its 2025 amendments; in a school setting it generally runs through the district's consent rather than a parent's. On top of both, most states now have their own student-privacy statutes, and many districts require a signed data privacy agreement before a single teacher logs in.
A vendor either has the paperwork and the third-party verification to stand behind those obligations, or it does not. You should not have to take anyone's word for it, ours included.
Three independent verifications and one public commitment, in plain language
Most vendors blur those two. A verification means an outside party examined us and issued a finding. A commitment means we signed something in public and can be held to it. Both are worth having. They are not the same thing, and a vendor that presents a pledge as though it were an audit is telling you something about how it handles the rest of its claims.
1. Google OAuth verification, including the restricted Google Drive scope
When ClassLens connects to your Google Classroom, it does so through Google's OAuth system. Google reviews the apps that request sensitive access, and it saves its most rigorous review for what it calls "restricted" scopes, a category that includes Google Drive.
ClassLens completed Google's full OAuth verification, including for a restricted Google Drive scope. Google itself examined how we request, use, and protect that access before it approved us. A lot of tools avoid the restricted-scope review because it is demanding. We went through it.
What it means for you: the connection between ClassLens and your Google environment has already passed Google's own security review.
2. CASA Tier 2 assessment
CASA is the Cloud Application Security Assessment, an industry framework for testing how an application handles security. Tier 2 is an assessment performed by an authorized third party, not a self-graded checklist.
ClassLens completed the CASA Tier 2 security assessment. TAC Security submitted the Letter of Validation on April 1, 2026, through the Google-authorized ESOF AppSec platform.
An outside security firm went through the application against the CASA requirements and validated the result to Google.
3. SOC 2 Type I attested
SOC 2 is the report district IT teams ask for most often. It is an examination of a service organization's security controls by a licensed CPA firm. Two things matter about how a vendor cites it, and many blur both.
ClassLens is SOC 2 Type I attested by Percilchofe CPA LLC (License No. 1188), with an unqualified opinion. "Attested" is the right word, because SOC 2 is an attestation by an auditor, not a pass-or-fail certification. "Type I" means the auditor examined our controls as designed at a point in time. Type II, which we have not performed, examines those controls over an operating period. We name our auditor and we state the type, because a vendor that says only "SOC 2," with no type and no auditor named, is telling you less than it sounds.
A licensed CPA firm reviewed our security controls and issued a clean opinion. The full report is available under NDA.
4. CISA Secure by Design Pledge signatory (a commitment, not a verification)
The Student Privacy Pledge that many edtech vendors once pointed to was retired in 2025. When it was retired, the Future of Privacy Forum pointed vendors toward the SDPC National Data Privacy Agreement and CISA's Secure by Design program.
Evolved Academics, the company behind ClassLens, signed the CISA Secure by Design Pledge on April 28, 2026. Signing commits us to seven concrete security goals: multi-factor authentication, reducing whole classes of vulnerabilities, a published vulnerability disclosure policy, and transparency about security issues, among others. Signatories appear on a public CISA registry, so this is a claim you can check yourself instead of taking it on faith. Look for Evolved Academics, our company name, rather than ClassLens, the product.
CISA does not audit, verify, endorse, or certify its signatories, and we do not present the pledge as though it did. It is a public commitment on a government-maintained list, which makes it checkable and makes us accountable to it. That is worth something on its own, and a vendor who dresses a pledge up as an audit is counting on you not knowing the difference.
Where student work goes, and what happens to it
Everything above is process. This is the part where a security-minded teacher or CIO pushes hardest, and they are right to.
ClassLens runs its AI evaluation on Google Cloud Vertex AI, under the Google Cloud Data Processing Addendum, with Zero Data Retention. Google retains nothing from our grading calls, and student work is not used to train AI models.
The details are where vendors get vague, so here is the whole path. During a job, each submission file is staged in a private Google Cloud Storage bucket in a U.S. region for inference, then deleted, with a 24-hour bucket lifecycle policy as the backstop if a delete does not land. In Grade and Review mode, so you can read the student's actual work beside the draft grade, the submission files and the drafted grades and feedback are held in a short-lived review queue until you release them, with a 24-hour failsafe expiry if you walk away without finishing. Grades are written back into Google Classroom, where they already live. After that, what stays on our side is operational metadata, students referenced only by opaque Google Classroom identifiers rather than by name or email address, and the class-level knowledge gap report. All of it is governed by the retention schedule we publish on our security page.
What we deliberately do not build
Look at what is missing from that path. ClassLens has no student profile, no per-student history, and no analytics that follow a student from one assignment to the next. That is architecture rather than restraint: there is no student name, email, or student ID column anywhere in our database. The per-submission ledger keys on an opaque Google Classroom submission identifier, and the knowledge gap report stores counts per rubric criterion, not rows per child. Building longitudinal student analytics on top of that would mean first collecting data we deliberately never collect.
Your students are already tracked over time in the systems built for it: Google Classroom and your student information system, whether that is PowerSchool, Infinite Campus, Aeries, or whatever your district runs. Those are the systems of record, with your district's retention schedule and your district's contracts behind them. We saw no reason to build a third copy.
So when you ask other vendors what their retention policy is, read the answer closely. "As long as necessary" is the common one, and it is a legitimate answer, but it describes a different shape of system than a window measured in hours. Ask what the number is.
And a teacher holds the release. ClassLens writes every AI-generated grade as a draft; putting it in front of a student is a separate, explicit teacher action. We permanently removed automatic return of grades in April 2026, so there is no mode in which a grade reaches a student without that action. We say it that way, rather than promising that every grade is read closely, because the release is the part we can actually guarantee in code. Whether you read every draft is your call, and the review dashboard is built to make reading them fast.
We back the grading itself with published data, too. Across more than 200 experiments on 41,000-plus student submissions drawn from eight published academic datasets, ClassLens agreed with trained human raters at a Quadratic Weighted Kappa of 0.623 on the ELLIPSE benchmark, slightly above the 0.612 agreement measured between the two human raters scoring the same essays. On a balanced 800-essay sample, demographic-bias effect sizes across English-learner status, gender, and race all came in below the 0.20 threshold researchers use to flag a concern. Security and accuracy are separate questions, but a vendor should be willing to show data on both. The full write-up is in our accuracy and bias testing post.
The questions to ask any vendor
You do not have to be a security expert to vet an AI-assistive grading tool. You have to ask six questions and expect specific answers. Run these on any vendor you are weighing, ClassLens included.
- Ask whether they will execute your district's data privacy agreement, and ask to see their vendor-side exhibits. A completed exhibit you can read is different from a marketing claim. If a vendor cites "customers," ask whether those are signed districts or individual teacher accounts.
- Ask for the SOC 2 report, then ask two follow-ups. Is it Type I or Type II, and which firm attested it? The words "SOC 2," with no type and no auditor named, are not an answer.
- Ask whether their Google OAuth is restricted-scope and fully verified. Some tools request only lightweight scopes to skip Google's harder review. Ask which scopes they use and whether Google verified them.
- Ask whether they signed the CISA Secure by Design Pledge, and under what legal name. Then check the public signatory registry yourself. This one takes about a minute to confirm, and the legal name matters, because it is often the parent company rather than the product.
- Ask which AI model processes student work, where it runs, and what the retention policy is. "We use AI" is not specific enough. You want the provider, the region, a yes-or-no on whether student work trains models, and a number for the retention window rather than an adjective.
- Ask whether a teacher reviews every grade before students see it. Human-in-the-loop is not a default you should assume. Ask for it in writing.
Few tools in this category can answer all six with specifics. The checklist moves the conversation from adjectives to evidence, and a vendor with real answers will be glad you asked. How they react tells you about as much as the answers do.
FAQ
How does ClassLens handle FERPA?
ClassLens operates under FERPA's school-official designation, which a district makes, and is COPPA-aligned, including the 2025 amendments. We will execute a district's NDPA or DPA, and our vendor-side exhibits are drafted and ready for your review. The substance behind those words is the data posture above: student work is processed transiently on Google Cloud Vertex AI, and nothing beyond operational metadata and the class-level knowledge gap report persists once a grading job is released.
Where is student work processed, and is it used to train AI?
On Google Cloud Vertex AI, under the Google Cloud Data Processing Addendum, with Zero Data Retention, which means Google retains nothing from our grading calls and student work is not used to train AI models. On our side, submission files are held only for the life of the grading job and its review window, with a 24-hour failsafe expiry.
What is the difference between SOC 2 Type I and Type II?
Type I examines whether security controls are properly designed at a point in time. Type II examines whether those controls operate effectively over a period, usually several months. ClassLens is SOC 2 Type I attested by Percilchofe CPA LLC, with an unqualified opinion. We state the type and name the firm rather than using a vaguer label, because the type and the auditor are what the report actually turns on.
Does a teacher review grades before students see them?
Every AI-generated grade is written as a draft, and no grade reaches a student until a teacher takes an explicit release action. Automatic return of grades was permanently removed in April 2026, so there is no setting that skips that step.
Can we get your compliance documents for a district review?
Yes. Our security overview and CoSN K-12CVAT responses are available on request, and the full SOC 2 Type I report is available under NDA. Start at our security page.
See it for yourself
If you are evaluating tools for your classroom or your district, read the details on our security page, where every claim above links to its source. Then try the product without a login on our demo.
If you are looking at this for a school or district, a short conversation is usually faster than a document exchange. Pricing for districts and individual teachers is on our pricing page, and the security page has a way to start a district conversation. Ask us the six questions above. We answer them the same way every time.