You handed in a paper you actually wrote. Days later, you are sitting in your professor's office being told an AI detector says you cheated. Your stomach drops. You have no idea how to prove a negative.

If this is happening to you — or to someone you love — you are not alone, and you are not without options. In February 2026, a New York court reversed exactly this kind of accusation and ordered the university to expunge the student's record. The legal and institutional ground has shifted. The single most important thing you can do right now is gather evidence, stay calm, and refuse to settle.

This is the playbook.

The new reality after Newby v. Adelphi

For most of 2023–2025, students accused of using AI faced a brutal asymmetry. The detector spat out a number — 87%, 96%, "100% AI-generated" — and that number functioned, in practice, like a verdict. Appeals were rare. Reversals were rarer. The burden of proof effectively flipped onto the student, who was asked to prove a negative against a piece of software the institution itself often did not fully understand.

That asymmetry began breaking in early 2026.

The case that broke it is Matter of Newby v. Adelphi University. Orion Newby, a first-year student at Adelphi in Garden City, New York, submitted a paper on Christianity and Islam for a World Civilizations course in Fall 2024. His professor ran it through Turnitin's AI detector. The tool flagged the paper as 100% AI-generated. Newby — who has documented learning differences and worked extensively with tutors through Adelphi's Bridges support program — denied using generative AI and explained that he had used Grammarly to help with grammar. He ran the same paper through two other AI detectors. Both classified it as human-written. Adelphi rejected his appeal anyway, with the same administrator presiding over both the original determination and the appeal.

On January 28, 2026, New York State Supreme Court Justice Randy Sue Marber ruled in Newby's favor and ordered Adelphi to vacate the finding. The judge held that the university had failed to provide a "meaningful opportunity to be heard," had refused to consider exculpatory evidence (the two other detector reports), and that an appeal process in which the same person decides and re-decides the case is not a meaningful appeal at all.

The decision is the first of its kind, but it will not be the last. Cases are now active against Yale, the University of Minnesota, and the University of Michigan. The legal direction of travel is clear: "the detector said so" is no longer an institutionally — or legally — defensible standard.

That is the door you are walking through. Here is how to walk through it.

The six forms of evidence that actually exonerate you

AI detectors produce probabilities. Probabilities are not proof. What is proof is the trail of writing you left behind while doing the work. The six forms of evidence below are what investigators, ombudspersons, and judges find persuasive — in roughly the order you should reach for them.

1. Drafts and earlier versions

Any version of the paper that pre-dates the submitted file is gold. Email drafts to yourself. Files in Dropbox, OneDrive, or iCloud with timestamps. Earlier .docx versions saved under different filenames. A messy first draft with bad sentences, abandoned arguments, and grammar mistakes is, paradoxically, your strongest defense — because AI does not produce abandoned arguments.

2. Google Docs (or Word) version history

If you wrote your paper in Google Docs, open it now and click File → Version history → See version history. You will see a sidebar showing every save, edit, and revision, often minute-by-minute, with the words you added or deleted highlighted. Microsoft Word's "Track Changes" and the version history in OneDrive offer similar trails. Do not edit the file again before preserving this. A clean, organic edit history — pauses, rewrites, sentence-by-sentence growth — is the closest thing to a forensic fingerprint of human authorship.

3. Browser and search history

Your Google search history, your YouTube history, the tabs you had open while researching — all of it can corroborate the timeline of your research. If you spent two hours on Wikipedia and JSTOR the night before the paper was due, that pattern is consistent with a student actually researching. AI-generated essays leave no such trail.

4. Scratch notes, annotations, and physical artifacts

Handwritten notes, annotated PDFs of source readings, sticky notes on library books, photos of your whiteboard, voice memos where you talked through an argument. These are the kind of human residue that AI doesn't produce. Photograph them. Keep them.

5. An oral defense of your own work

If you wrote the paper, you can talk about it. Volunteer — proactively — to sit down with your professor and explain your argument, your sources, your choices, the parts you cut. This is increasingly considered the gold standard for resolving authorship disputes, because it is the one form of "evidence" that AI cannot fake on your behalf. Ask for it. Insist on it.

6. Process recordings

The newest and most powerful form of evidence: a verifiable, timestamped recording of how the paper was actually written — keystrokes, pauses, copy-paste events, and revisions. Tools designed for this (including Mentafy's Authorship Report) generate a tamper-evident record of the writing process itself. If you have one, you have something no detector can argue with. If you don't — read on for how to start.

How to read your own detector report (and spot the weak findings)

Before your hearing, request a copy of the detector report. Then look for the specific patterns that have been documented as unreliable:

• Short text inflates false positives. Most AI detectors lose accuracy dramatically on passages under 300 words. If the flag is on a single paragraph, it is structurally weak evidence.
• Formal, simple prose gets flagged. AI detectors learn to flag low "perplexity" — text that uses predictable word choices and common sentence structures. That is also the writing pattern of careful students, scientific writing, and non-native English speakers. A 2023 Stanford study by James Zou and colleagues found that seven leading AI detectors falsely flagged 61% of TOEFL essays by non-native English speakers as AI-generated, and 19% of those essays were unanimously misclassified by every detector tested.
• Grammarly and similar tools cause artifacts. Running your paper through Grammarly, Microsoft Editor, or any grammar checker can shift sentence structure in ways that look "AI-like" to a detector. This is exactly what happened in the Newby case. If you used a grammar tool, document which one and when.
• Bias in the underlying model. A 2024 Common Sense Media survey of more than 1,000 American teens found that Black students reported being falsely accused of using AI at roughly twice the rate of their white peers (20% versus 7%). If you belong to a demographic group that detectors are known to flag disproportionately, that pattern is itself part of your defense.
• Cross-detector disagreement. Run the same paper through 2–3 other detectors (GPTZero, Copyleaks, ZeroGPT, Originality.ai). If they disagree, you have a Newby-style argument: the technology itself does not agree with itself. The OECD and multiple peer-reviewed studies have shown no detector exceeds reliable accuracy thresholds.

The first 48 hours: a step-by-step playbook

What you do in the two days after an accusation matters more than almost anything else. In order:

1. Do not panic-edit your files. Do not open the original document and start changing things. Do not "clean up" your drafts. Do not delete browser history. Preserve everything as it currently exists.
2. Preserve your version history immediately. Open Google Docs, take screenshots of the version history panel, and export the full file. For Word, copy the entire OneDrive folder. Email these to yourself with a timestamp.
3. Request the detector report in writing. Email your professor (politely) asking for the full report, the detector used, and the date it was run. Keep it in writing — verbal accusations and conversations vanish; emails don't.
4. Do not confess to anything you didn't do. Under pressure, students often agree to a "lesser charge" just to make the meeting end. Don't. Many institutional integrity systems treat any admission as final and use it against you on any second accusation.
5. Get a second and third detector opinion. Run your paper through 2–3 alternative AI detectors. Save the results as PDFs. If they conflict with the original — and they very often do — that contradiction is now part of your file.
6. Identify a support person. Your institution likely has an ombudsperson, a dean of students, a writing center director, a disability services office, or a student advocate. Find them today. You are entitled to bring a support person to most hearings; check your code of conduct.
7. Read your institution's academic integrity policy from front to back. Print it. Highlight every procedural right it gives you — notice requirements, evidence standards, appeal rights, hearing rights. Then check whether the institution is actually following its own rules. In Newby, the procedural failure mattered more than any technical argument.
8. Volunteer for an oral defense. Email your professor and offer to walk through your paper, your sources, and your argument in person. Frame it as a request, not a concession.

What your institution's process should look like — and the red flags

A defensible academic integrity process — the kind that survives an appeal in 2026 — looks like this: written notice of the specific accusation, access to the evidence being used against you, a reasonable amount of time to prepare a response, a hearing in front of someone independent of the accusing professor, the right to bring a support person, the right to submit your own evidence, and a meaningful appeal to a different decision-maker than the original.

The red flags — the things that Newby tells you to look for — include any of the following: the only "evidence" against you is a single detector output; the professor will not show you the report; the same person hears your appeal as decided your case; you are not allowed to bring a support person; you are pressured to "just sign" a finding to make it go away; your accommodations (IEP, 504, disability services) are not being considered; the process is described as "non-disciplinary" but the consequences (zero on the paper, mandatory remediation, second-offense escalation) are clearly disciplinary.

If you see two or more of these, you are looking at a process that may not hold up — and that is a fact worth documenting.

Proactive insurance: the Authorship Report as a credential of originality

Here is the larger shift that the post-Adelphi moment is making possible. For two years, the conversation about AI in education has been framed as surveillance: institutions surveil students, detectors surveil text, students surveil their own writing for "AI-sounding" patterns. It is exhausting, adversarial, and — as the data keeps showing — biased.

There is a better frame. Treat your authorship as a credential you can voluntarily demonstrate, the same way you might attach a transcript to a job application. A Mentafy Authorship Report is a tamper-evident, timestamped record of how a piece of writing came into existence — the pauses, the revisions, the abandoned drafts, the moments you got stuck. It is generated with you, not to you. You choose when to share it, and with whom.

The students who will fare best in the next three years are not the ones who get better at evading detectors. They are the ones who can prove their work when asked — and increasingly, who don't wait to be asked. Attaching an Authorship Report to a high-stakes paper before submitting it does to a false accusation what a receipt does to a returned package: it ends the argument before it starts.

That is what we mean when we talk about flipping integrity from gotcha to credential.

When to call in reinforcements: lawyers, ombudspersons, and disability services

Most accusations can and should be resolved inside the institution. But there are clear points at which outside or specialized help becomes essential.

Talk to your institution's ombudsperson as soon as you receive the accusation. Their job is to be neutral and confidential, and they will know which procedural rights you actually have at your school. This costs nothing and creates no record.

Contact disability services if you have any documented learning, neurological, or language difference — or if you used assistive tools (text-to-speech, Grammarly, dictation, translation, structured writing supports). The Newby case turned, in part, on Adelphi's failure to weigh his documented disability and the tutoring support he had received through the university's own program. If your accommodations were not considered in your case, that is a serious procedural problem.

Contact an education attorney if any of the following are true: the penalty involves suspension, expulsion, or a permanent transcript notation; you have already exhausted internal appeals; the institution is refusing to follow its own published procedures; or your case involves an international student visa (where any disciplinary finding can have immigration consequences). Many education attorneys offer free initial consultations. Some state bar associations maintain referral lines. Cases like Newby have established that the legal route is no longer a long-shot.

Connect with student advocacy organizations such as FIRE (the Foundation for Individual Rights and Expression), which has begun tracking AI accusation cases, and your campus student government's academic affairs committee, which may have a student defender program.

The bottom line

You did not invent the technology that flagged you, and you do not have to accept its verdict. The Adelphi ruling, the FTC's 2024 settlement with the detector company Workado over inflated accuracy claims, the Stanford and Common Sense Media studies on bias — all of it adds up to one thing: the era of "the detector said so" is ending. What replaces it is a system in which the writing process itself becomes the evidence, and the student becomes the credential-holder rather than the suspect.

If you are facing an accusation today, work through the steps in this playbook. If you are not — start protecting your authorship now, before you ever need to. Either way, the most important thing you can carry into the next conversation is this: you have rights, you have evidence, and the ground is shifting under your feet in your favor.

---

Further reading on Mentafy

• AI Detection Tools: When You Turn It In, It's Too Late — why submission-based detection structurally cannot tell who wrote what.
• How AI 'Humanizer' Tools Are Making Cheating Invisible — the arms race that proves why process evidence is the only durable answer.
• The Mentafy Reference Verifier and Authorship Report — the tools behind the playbook.