Somewhere in the last year, AI grading went from "weird experiment" to "something your department chair might ask about." That's a real shift, and it deserves a real conversation, not a vendor pitch.
I built ClassLens as a 22-year classroom teacher. I use it on my own students' work. So this isn't an abstract ethics post. These are the rules I actually grade by, and the rules I think every teacher should grade by, whether they use ClassLens, a competitor, or a chatbot they opened on a Sunday night.
Responsible AI grading lives on two pillars: student privacy and student connection. If you respect both, AI becomes a real assistant. If you compromise either one, you've crossed a line you shouldn't cross.
A Story From My Own Classroom
Let me tell you the story that shapes how I think about this.
After about a year of fine-tuning ClassLens, I was grading résumés for my CTE and Engineering Design class, by hand. If you teach in career pathways, you know the drill: résumé workshops are an every-year assignment. It's a norm in CTE because a résumé isn't a classroom exercise. It's the thing a student will actually hand to a real employer a few months after they leave your room.
That's why I grade résumés personally. I like to read them. You learn a lot about a student from a résumé: what jobs they've already held, what they volunteer for, who they are outside your classroom. And I want them to walk out of my class with something that will land an interview, not something that will get them screened out.
I got to a résumé from one of my top students. 4.38 GPA. The formatting was sharp. The content was strong. I skimmed through it, recognized the quality I expected from her, and gave it a 10 out of 10.
Later, I ran ClassLens over the whole class as a calibration check. I do that throughout the year to make sure my grading stays consistent across a hundred-plus submissions. When I got to her résumé in the dashboard, ClassLens had given it a 9 out of 10.
I read the comment. She had misspelled the word "present" six times.
I went back to the résumé. Sure enough, my eye had slid right past every one of them. Six misspellings of a common word, on a résumé I gave full credit to.
If I hadn't run ClassLens, she would have submitted that résumé to a real employer thinking, "My teacher gave me a 10 out of 10. This is ready." It wasn't ready.
That's the case for using AI responsibly on the right kinds of work. It catches the details you want to catch but can't always, because you're tired, because you know the student and your brain is filling in what it expects to see, because the formatting is nice and you trust your read of it. It's especially valuable on the assignments where you'd love to point every issue out to a student but you don't have the time, and you end up giving full credit or no credit just because you couldn't get to the details.
On the right kinds of work, AI doesn't replace the teacher. It covers what the teacher missed.
The rest of this post is about telling those "right kinds of work" apart from the ones you should never hand to a machine.
Pillar 1: Privacy
Student work is student data. The moment you paste an essay into a tool, you've shared something that belongs to a minor, usually without their informed consent and often without their parents' knowledge. That's a serious thing. Here are the rules.
Rule 1: Never use a tool that trains its models on student work
Most "free" and bargain-tier AI tools pay for themselves by using your inputs as training data. That means the essay a 14-year-old wrote about their grandmother's immigration story can end up, in fragments, inside the next version of a commercial model.
Before you upload a single piece of student writing, find the answer to this question on the vendor's website: "Is my data used to train your models?" If the answer is yes, or the answer is hidden behind three layers of legalese, or the answer only applies to "enterprise customers," walk away. Student work is not training data. It's a child's thinking, and it doesn't belong in a corporate dataset.
Rule 2: Never use a platform that stores or tracks your students
The second question to ask: "What do you do with the data after you process it?"
A responsible AI grading tool processes the submission, returns the grade, and doesn't retain the student's work, identity, or behavior as a long-term record on its servers. It doesn't build profiles. It doesn't track which students "write the weakest essays" or "struggle most with thesis statements" in a way that follows them beyond your classroom.
If the tool markets "student analytics" that persist across assignments and across years, ask hard questions. Whose intelligence is being accumulated, and for whose benefit? The teacher's, to improve instruction? Or the vendor's, to package and sell?
Rule 3: Trust your district's vetting, or do your own homework
Here's the part I want to be honest about. If your district adopted ClassLens, or any AI tool, through its normal procurement process, you don't actually have to do this work yourself. Your IT and curriculum teams already asked the questions that matter. They reviewed the vendor's privacy policy. They signed a data privacy agreement, or at minimum looked one over. They have a record of what data is going where. That's what having a district stack is for. Lean on it.
Where it gets risky is the side door. A teacher tries a free tool on a personal device, with their own Gmail, on student work copied out of Classroom. Most experimentation in education starts that way, and there's nothing wrong with experimenting. But the moment you go around the district's vetting process, the privacy decision lands entirely on you.
If that's where you are, here's the short list to check before you upload student work:
- Can you find a privacy policy in two clicks? If not, that's the answer.
- Does it say in plain English that the company won't use your data, or your students' work, to train its models?
- Is there a contact email for privacy questions? Real companies have one. Side projects often don't.
- Does the tool offer schools a data privacy agreement, and does it claim FERPA and COPPA compliance? That's the floor, not the ceiling.
You don't have to be a lawyer. You just have to know the basics. And you have to know that "free and easy to sign up for" usually means the company is making money some other way, and often that other way is your students.
Pillar 2: Connection
Privacy is about what you do with a student's work. Connection is about whether the machine was ever supposed to read that work in the first place.
Some assignments are a transaction: here's a prompt, here's a rubric, here's what "proficient" looks like. An AI assistant handling those well, against your rubric, in your voice, under your review, is a gift to a teacher's weekend.
Other assignments are a relationship. They're the student showing you something about themselves, and they need to be seen by a human. Full stop.
Rule 4: Never use AI to grade identity work
"All about me" essays. "Where I come from" narratives. Personal-statement drafts. Reflection pieces where a student is telling you, often for the first time, something real about their life.
When a 7th grader writes about the custody battle, the grandmother who died, the country they left, the language they're trying not to lose, the most important moment in the entire assignment is the moment a human teacher reads it and responds as a human. A machine responding to that essay, even gently, even with encouragement, sends the wrong message. Your story went to a robot. The robot had thoughts.
Grade those yourself. Every time. No exceptions.
Rule 5: Never use AI to grade subjective, creative, or hands-on work
Art. Engineering design projects. Lab notebooks from a week-long build. Sculpture. Performance. A student's photography portfolio. A robotics team's iteration log. A capstone project.
These assignments don't have a clean rubric that maps to the visible output. They have a rubric that maps to the process: the iteration, the creativity, the craft decisions, the student's growth from draft 1 to draft 4. A language model reading a PDF of the final product can't see any of that. It can't smell the glue. It can't watch the student fail and recover. It can't recognize, in a single sentence of a lab notebook, the exact moment the student finally understood what was happening.
Machines are good at pattern-matching against explicit criteria. They are bad at recognizing a teenager becoming good at something. Those assignments, the ones where the real grade lives in what the student couldn't see in themselves, belong to you.
If you do choose to use AI on a borderline assignment
Some assignments sit in the middle. A personal essay with a clear rubric. A design project with rubric-scorable written components. A lab report attached to a hands-on build.
If you choose to use AI assist on something like that, two rules:
- Grade only the rubric-scorable parts with the AI. Let it evaluate the clarity of the writing, the citation format, the presence of required sections. Flag everything subjective for your own eyes.
- Be honest with your students.Tell them, "An AI assistant helped me score the rubric portions, and I personally read the reflective sections." Students can handle that truth. What they can't handle, and shouldn't have to, is finding out later that the feedback they got on their most personal work came from a machine that was pretending to be you.
The Through-Line: The Teacher Is Always in the Loop
If you take one thing from this post, take this: AI grading should never take the teacher out of the loop. It should just take some of the load off, in the places where it's safe to do that.
That's why ClassLens, by design, never releases grades to students without the teacher reviewing them first. Every grade lands in a dashboard. You review every one. You release the batch when you're ready. There is no mode, no setting, no toggle that lets the AI send feedback straight to a student unseen by you. We removed that capability from the product on purpose.
The ethics here aren't complicated. Students deserve privacy. Students deserve connection. Teachers remain accountable for what their classroom tells a student about who they are.
If a tool asks you to compromise any of that for speed, the tool isn't worth the time it saves.
Try It
If you're ready to try AI grading in a way that protects student privacy, never trains on student work, and keeps you as the final reviewer on every grade, ClassLens was built for exactly that.