Packback

It’s an exhausting position to be in: looking at a student submission and wondering if a human actually wrote it.

But a January 2026 survey of college students reveals that your students are experiencing a parallel exhaustion along with a collapsing trust that has become almost as ubiquitous as AI.

82% of students report they do not use AI to write their assignments. Instead, they use it as a digital sounding board to brainstorm, outline, or clarify difficult material. Yet two-thirds believe their peers are using it inappropriately. And nearly 3 in 4 now fear being falsely accused – even if they didn’t use AI at all!

Scenarios like this are why educators have quickly realized that policing AI is a losing game that only serves to erode classroom trust. So if detection isn't working, what do we do next? We must shift from grading the artifact to tracking the process of learning itself. Our report, The AI Trust Gap in Higher Education, breaks down the data and shows how institutions are moving past policing toward making student thinking visible. We'll drop the link in the comments.

The assumption that the majority of students are using generative AI to entirely bypass assignments does not hold up against student survey data. In fact, a recent Packback student survey revealed that 82% have never used AI to write a full assignment -- but they fear false-positive accusations regardless. The data explicitly shows that concern over false accusations is not significantly related to how frequently a student uses AI.

What students are experiencing instead is an institutional trust gap.

When we rely solely on black-box detection scores to police the final artifact, we introduce severe governance risks, elevate student complaints, and disrupt the faculty-student dynamic.

The path forward requires a structural shift away from reactive policing and toward Learning Intelligence. By instrumenting and documenting the actual authorship process (tracking outlines, drafts, revision timelines, and citation actions) we make student thinking visible. This eliminates the black box of submissions, protects authentic student effort, and removes the liability of inaccurate detection models.

We get into these findings and what to do next in our latest research report, "The AI Trust Gap in Higher Education." Grab your copy here and download the full higher education strategy framework to close the campus trust gap: https://hubs.ly/Q04pBXhN0

Students who want to cheat will always find a way, but the vast majority are just trying to figure out how to use AI meaningfully while still producing their own work. Watch our latest on-demand webinar with Kelsey Behringer to learn how to move beyond detection and start measuring the actual thinking process. Watch the on-demand and get a copy of the presentation slides and all resources shared during the session here: https://hubs.ly/Q04r09hn0

Trying to catch generative AI with a post-submission detector is like trying to catch the wind with a butterfly net.

Detection happens after the assignment is finished and by then, the thinking either happened or it didn’t.

Not only that, students deserve to be recognized for more than the final document they submit.

Their research matters.
Their revisions matter.
The questions they asked matter.
Their ability to challenge an AI-generated claim matters.
Their reflection on what worked and what failed matters.

If we want students to retain ownership of their work, we have to build courses that reward planning, iteration, verification, reflection, and judgment along the way.

That is how we move from catching cognitive surrender to preventing it.

Read more in our latest blog: From Detection to Prevention: Building Infrastructure for Cognitive Vigilance
https://hubs.ly/Q04q07820

👋 We're excited to be heading to InstructureCon 2026, kicking off this week in Louisville!

Join the Packback team as learning leaders, instructional designers, and educators come together to discuss pedagogy, academic integrity, and student success across the industry.

If you'll be attending, be sure to join us on 📅Thursday, July 23, at 2:30 PM in room M105 for a breakout session led by Packback CEO, Kelsey Behringer, "Packback: Building AI Literacy Through Better Writing, Discussion, and Feedback."

We’ll explore how institutions and faculty can help students engage more deeply, strengthen their writing process, and use AI responsibly without outsourcing their thinking. Attendees will leave with a practical framework for evaluating AI’s role in engagement, feedback, writing, and critical thinking while keeping people, pedagogy, and learning outcomes at the center.

Can't make the breakout? Be sure to stop by Booth #46 to meet our team and grab a copy of the framework!

AI strategy gets more concrete when you can see what it looks like in the learning workflow. Join us live on July 14 for a 30-minute platform tour and we’ll show you how our Learning Intelligence and Instructional AI helps make the writing process visible, reduces grading time, and supports student thinking without replacing the instructor. If you have been curious about Packback but want a no-pressure walkthrough with a chance to ask any questions you may have -- this is for you!

Mark your calendar and save your seat now:
https://hubs.ly/Q04mvPYh0

For the first time in 20 years, Professor Roberto Serrano of Brown University gave his students a take-home midterm this Spring since a gunman had killed two students on campus in December. Many students were afraid to be in a classroom and giving them the exam at home was the compassionate call.

He made the exam harder than usual, figuring unlimited time warranted a bigger challenge.

The average midterm grade? 96%.
His historical midterm averages ranged from 65-80%.

He told the class of 86 that he suspected AI use and switched the final to in-person. He gave them a direct message: if your scores hold up, I'll count the midterm.

18 students dropped the class.
9 enrolled but didn't take the final.

The final exam average: 48.6%.
19 students failed.

He submitted the data to Brown's academic integrity committee in May but they didn't respond. He went public in June and the committee then asked him to file individual complaints for the majority of his 86-person class.

He called it "ridiculous" and hundreds of Brown alumni emailed him in support.

Here is what we keep coming back to.

Serrano did everything right. He extended grace after a campus tragedy. He noticed something was wrong and investigated. He redesigned his assessment mid-semester. He warned students and gave them a chance to prove him wrong. He escalated through the proper channels. And the institution handed him a stack of paperwork.

The problem is not that professors don't care about academic integrity. They do. The problem is that we keep handing them a detection challenge when what they are actually facing is a design assessment challenge.

AI did not break Serrano's exam. That exam was always one well-coached student or a circulated answer key away from being gamed. AI just scaled that vulnerability to 86 students overnight.

Faster detectors will not fix this. Stricter honor codes will not fix this.

What actually works is assessment design that makes the thinking process visible. Assignments where students plan, draft, revise, and explain their reasoning at each step. Structures where AI becomes a thinking partner students are required to critically evaluate, not a ghostwriter they can pass off as their own.

That kind of infrastructure is not yet widespread. It requires institutional investment in pedagogy, not just policy.

Until that changes, professors like Roberto Serrano will keep doing the right thing, doing it alone, and wondering why the system they trusted did not show up.

𝐀𝐈 𝐝𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐝𝐨𝐞𝐬𝐧'𝐭 𝐰𝐨𝐫𝐤, 𝐬𝐨 𝐰𝐡𝐚𝐭 𝐝𝐨 𝐰𝐞 𝐝𝐨 𝐢𝐧𝐬𝐭𝐞𝐚𝐝? When generative AI entered campuses, the first reaction across higher education was to try to control it. But post-submission detectors has turned faculty into full-time AI police and created an adversarial environment with students.

The reality is that AI use is here to stay, and it splits student behavior distinct directions depending on assignment design:

⛔𝐂𝐨𝐠𝐧𝐢𝐭𝐢𝐯𝐞 𝐒𝐮𝐫𝐫𝐞𝐧𝐝𝐞𝐫: Unstructured use where a student relinquishes intellectual control, inputs a prompt, and submits the chatbot's output as their own. The learning journey ends before it begins.

🔬𝐂𝐨𝐠𝐧𝐢𝐭𝐢𝐯𝐞 𝐎𝐟𝐟𝐥𝐨𝐚𝐝𝐢𝐧𝐠: Guided integration where lower-level tasks are intentionally delegated to AI, freeing up cognitive resources for higher-order critical thinking, judgment, and explanation.

🤔𝐂𝐨𝐠𝐧𝐢𝐭𝐢𝐯𝐞 𝐕𝐢𝐠𝐢𝐥𝐚𝐧𝐜𝐞: Only when cognitive offloading is accepted and intentionally guided can it develop into Cognitive Vigilance -- the active, systematic scrutiny of AI output where students deepen their understanding by auditing claims, verifying reasoning, and cross-referencing sources.

If we only grade the final artifact, we cannot tell these paths apart. It is time to shift our focus from policing the final essay to instrumenting and validating the writing process itself.

Read the full breakdown on how to move your course design past the detection cat-and-mouse game: https://hubs.ly/Q04p41T50

Remember how your middle school math teacher always demanded that you "show your work?" They cared more about seeing what steps you took than seeing the right answer circled at the bottom of the page.

In the age of GenAI, higher ed has to start treating writing and critical thinking the exact same way.

When we only grade the final, polished essay, we are practically begging for an AI crisis. A perfect output doesn't tell us if a student wrestled with the concepts, revised their thinking, or actually learned anything. And relying on AI detectors to figure it out for us is a losing game.

We need visibility into the learning process, not stricter AI policies.

See why Human-Centered AI is the only sustainable way to protect academic integrity in higher education: https://hubs.ly/Q04lGvLJ0

We need to talk about the collateral damage of AI detectors.

Right now, real students (especially neurodivergent learners and non-native English speakers) are getting flagged for cheating on essays they wrote completely from scratch.

The emotional toll of a false positive is devastating for the student, and the anxiety of making that false accusation is just as heavy for the instructor. We are relying on fragile algorithms that guess based on word predictability, and it is destroying the trust in our classrooms.

Integrity can no longer be defined as simply "the absence of AI."

We wrote a piece on why the higher ed integrity crisis cannot be solved by better detectors, and what a human-centered alternative actually looks like. Let's stop guessing and start focusing on the actual climb.

Read it here:

Why Trying to 'Catch' AI Use is a Losing Game for Higher Education Relying on AI detection damages trust and yields false positives. The real solution is grading the learning process.