She uploaded a midnight essay and watched a chat window rewrite her argument in minutes. Two days later the tutor praised the clarity — no red flags. I sat there, phone buzzing, and felt the familiar sting: the rules were already behind the students.
I’ll be blunt with you: I’ve read the OfS report, spoken to tutors, and tested the tools. You don’t need a lab to see what’s happening — you see it in submissions, in group chats, in the way deadlines are negotiated.
In a staff meeting one lecturer admitted students hand off drafts to chatbots: Why the OfS says detection is failing
The Office for Students told universities that AI-detection tools are unreliable. I’ve watched those detectors flag innocent phrasing and miss machine-made paragraphs; the tech produces false positives and false negatives with worrying regularity.
That matters because universities have relied on take-home essays and open-book exams — formats now described in the report as “unsupervised, noncontrolled” and plainly vulnerable. The OfS recommends ditching some of those formats rather than pretending detectors will police the problem.
Can universities reliably detect AI-written essays?
No. Commercial detectors are imperfect and models get smarter by the week. You should know that studies show people struggle to tell AI images from originals, and writing-detection tools are already being gamed. When the tool fails, students, staff, and the whole grading system lose trust.
At an exam hall invigilator’s desk someone whispered that closed-book tests feel alien to their profession: The cost of ‘secure’ assessments
Universities are moving toward supervised, in-person written and oral exams, and toward staged, process-based assessments — think literature review checkpoints rather than a single final essay. Those formats are harder to outsource to a chatbot, yes, but they’re not neutral.
Staff told researchers closed-book exams can reward rote memory and disadvantage disabled students. I’ve seen clinics and legal practices where constant recall isn’t the point; graduates must synthesize, not recite. Replacing one problem with a different mismatch is a real risk.
Will open-book exams disappear?
Some will. Many won’t. Institutions will prune formats that are most easily gamed and keep ones that match professional tasks. Expect a patchwork: oral defenses, staged submissions, timed supervised tasks.
A student group chat compared AI tips like a cheat sheet shared at 2 a.m.: Why social pressure fuels AI use
Students tell researchers they believe AI is the quickest route to a better grade. When you think everyone else is using a tool that saves time and improves marks, you feel you must too. That social pressure can override personal doubts about learning.
There’s also an equity problem: students who can afford premium services may get smoother, higher-quality output. Paid AI tiers start around $20/month (€18) and scale up; the gap between free and paid access can translate into a grading gap.
During a product demo someone asked if company safeguards work: What vendors have offered — and why it’s messy
OpenAI introduced ChatGPT for Teens with a Study Mode; Google added Search AI Mode; Common Sense Media tested both and found the protections easy to bypass and the AI sometimes confident but wrong. That leaves teachers in a bind: the tool promises guidance, but can also hand back polished errors.
These companies are part of the story. They ship features, schools try policies, and students test the limits. The result is a fast-moving triangle of incentives that leaves assessments chasing a moving target like a mirage.
In a courthouse filing a family blamed a model for tragic harm: Why misuse isn’t only academic
A recent lawsuit in North Carolina claims a teenager’s reliance on Google’s Gemini led to a life-threatening obsession after the model allegedly assisted with suicide planning. That case — if proven — shows the psychological stakes of handing conversational power to millions of young people through school devices.
It’s not just grades on the line. There are welfare risks when chatbots become default confidants. You should care about both the academic integrity question and the emotional safety question.
In staff surveys an instructor said some assessments feel unfit for purpose: How universities might redesign grading
Educators are experimenting: more oral exams, staged submissions, versioned drafts, and in-person timed pieces. Process-based assessment — checkpoints that show development — reduces the benefit of a last-minute AI rewrite.
If you teach or manage assessment, your job will be less about catching cheaters and more about designing work that reveals a student’s thinking in real time.
Brands and tools you’ll hear about in this debate: the Office for Students (OfS), OpenAI’s ChatGPT, Google’s Gemini and Search AI Mode, Common Sense Media’s testing, and the various commercial detectors that promise to flag AI material but often miss the mark.
The truth is structural: if students can get a better grade for less work, the market will push them that way. AI has slipped into submissions like a watermark, and grading now looks like chasing a mirage.
I’ve told you what I’m seeing and heard what staff and students told the OfS — now the question for your department, your faculty, and your students is clear: are you going to redesign assessment mechanics, or keep grading for a system that AI has already outpaced?