Over the past year, several well-known universities have restricted or disabled AI detection tools. Yale moved detection to informal screening only, so that a score cannot be cited in a formal complaint. Johns Hopkins made its tool advisory, meaning a flag can start a conversation but not a charge, and Waterloo turned off its detector entirely after internal testing flagged wholly human-written work as machine generated. At least a dozen institutions have taken some version of this step.

The reasoning is straightforward. Reported false-positive rates for human writing cluster somewhere between the mid-teens and mid-twenties in percentage terms in some independent tests, which is far too high for any process that can end in a misconduct finding. Accusing an honest student carries serious consequences for their record, their confidence, and their relationship with the institution. Departments that relied on a score are now looking for something sturdier.

The alternative is not to give up on integrity. It is to build a process that rests on evidence a person can examine and a student can respond to. That means designing assignments to reveal thinking, collecting process evidence, and talking with students. A department can adopt this approach without any new software, and many departments find that the shift reduces conflict with students because the process feels transparent and not secretive.

Redesign assessment so authorship is easier to see

Start by looking at which assignments are most exposed. A generic five-page essay on a broad topic is far easier to outsource than a paper tied to specific class discussions, local data, or a student's own earlier drafts. Add staged deliverables such as a proposal, an annotated source list, and a draft with comments. These steps give instructors a trail that shows how the final product developed.

• Treat detector scores as advisory at most and never as sole evidence
• Add staged deliverables such as a proposal, source list, and draft
• Use a short conversation about the paper as a routine verification step
• Apply the same verification standard to all students on major assignments
• Publish a department policy that lists what other evidence will be considered

Use short conversations as the verification step

A five-minute conversation about a paper is often the strongest evidence available. Ask the student to explain a key claim, defend a choice of source, or describe what they would revise. Students who wrote the work can usually do this comfortably, and the exchange leaves a record that either supports or raises concerns. Treat it as routine for major assignments so it does not feel like an accusation.

Be consistent about who is asked and why. If only students who seem unusual are called in, bias can creep into the process, particularly against multilingual writers. A random sample or a uniform requirement for all students is fairer. Document the questions asked and the answers given, and recording the reason each conversation was held makes it easy to show later that the standard was applied evenly across the class.

Write a department policy that states the limits of any tool

Put in writing that detector scores are not evidence of misconduct on their own and that no penalty will be based on a score alone. Specify what other evidence the department will consider, such as version history, notes, prior writing samples, and the student's explanation. Spell out how the student may respond and who reviews the case. A written policy protects students and faculty alike.

If your institution does use a detector, publish the false-positive rate you assume and what you will do when a flag appears. Share the policy with students at the start of each term so no one is surprised. Review it each year as tools and evidence evolve. Transparency builds the trust that integrity processes rely on, and if the assumed rate is high for a particular group of students, say so plainly and explain how those students are protected from unfair outcomes.

Support instructors who are doing the work

Moving away from scores asks more of instructors in the short term. Offer shared templates for staged assignments, a short guide for authorship conversations, and sample language for syllabi. Where permitted, rubric-based tools can reduce the time spent on first-pass grading so that instructors have more capacity for these conversations. The teacher still reads the work and makes every final decision.

Collect examples of assignments that have worked well and share them in a common folder. A department that learns together will develop stronger practices than one in which each instructor reinvents the wheel. Hold a brief meeting each term to discuss what is working. Over time, the process becomes part of the department's culture, and a quarterly one-hour working session is usually enough to keep the shared folder current and useful to new colleagues.