Detection is retreating.
The market is advancing.
Four stories that define the week of July 21–27.
Universities are quietly dropping AI-detection tools over accuracy fears, the FT reports — even as AI-related misconduct cases rise. An audit of 163 UK universities found more than 40% have no publicly accessible AI policy at all.
The Brown case goes national again. Economist Roberto Serrano takes his critique of case-by-case integrity procedures to a Chronicle interview; the take-home-vs-proctored gap (96% → 48%) remains the core evidence.
Anthropic launched Claude for Teachers as the big labs push free and cut-price education tools — with Coursera and Instructure partnerships. Early educator reviews: "no differentiator here."
Teachers in the building: zero. A teacherless, AI-taught private school opens in Tulsa this fall, and an upstate New York district will deploy a humanoid robot classroom aide named Sally.
The week's four stories
that matter for your campus
Selected for impact, novelty, actionability, evidence quality, and urgency.
Universities are quietly walking away from AI detection
The Financial Times reports that universities — the reporting centers on the UK — are dropping AI-detection software over accuracy concerns, even as AI-related misconduct cases rise inside internal procedures. False positives, which disproportionately flag non-native English speakers, are driving the retreat. An audit of 163 UK universities cited by the FT found more than 40% had no publicly accessible AI policy.
The chilling effect is measurable: in HEPI's December 2025 survey of 1,054 UK undergraduates (covered last issue), 42% said fear of being falsely accused made them less likely to use AI at all. The European Students' Union warns detection systems operate as closed boxes students cannot contest — one representative asks "whether higher education should rely on detection-based approaches at all."
Read the FT report →If your integrity process leans on a detector score, this is the week to ask what evidence would actually stand up in a misconduct hearing — and what replaces the detector when it's gone.
Update: Brown professor takes his AI-cheating critique national
Covered in the past two issues: Roberto Serrano's allegation of mass AI cheating on his ECON 1170 take-home midterm, and Brown's formal investigation. New this week: an extended Chronicle interview. "With AI, the cost of cheating has become zero," he says, renewing his criticism of per-student complaint filing: "We cannot afford to remain silent, even if it's unclear what to do." A companion Chronicle essay reports students describing AI use as compulsive.
The Chronicle interview →AI labs muscle in on the $6tn education market
Anthropic launched Claude for Teachers, free for US K-12 teachers; OpenAI shipped ChatGPT for Teachers in November; Google and Microsoft already have school versions. Both OpenAI and Anthropic pitch "Socratic" tutor modes. Worth noting: educator reviews are mixed — "It doesn't look any different than anything I've seen from OpenAI or from Google" — and the Socratic claims are positioning, not evidence.
FT analysis → Educator reactions →Where did all the computer-science professors go?
The Atlantic reports AI companies are hiring away top university researchers at a pace that has become "a punch line in academia" — "'I'm joining Anthropic' is the new meme right now." The poaching now extends past CS: a Stanford economist, a Maryland theoretical physicist, a UT Austin philosopher. Same-week context: the White House released a research strategy steering funding outside universities. Faculty retention in AI-adjacent fields is now open competition with industry labs — worth raising with your dean before your strongest colleagues become the meme.
The Atlantic → White House strategy →What the evidence says.
What practitioners are doing.
Distilled for a working professor — the takeaway, not the abstract.
From the Research
The grade-inflation panic may be outrunning the evidence
Studied: whether GenAI availability inflated grades in "GenAI-susceptible" courses (take-home essays and problem sets vs. in-class exams) at a large university — the "GenAI substitution hypothesis" (Zumel Dumlao et al., University of Michigan-affiliated team).
Found: in the authors' words, the findings "temper concerns that GenAI inflates grades and reduces students' satisfaction." The feared disproportionate grade rise did not clearly materialize. Preprint; not yet peer reviewed.
Use it: before overhauling assessments on the assumption of rampant inflation, check your own course-level data — panic and evidence may diverge.
arXiv preprint →Doctoral students are governing their own AI use — one professor at a time
Studied: how doctoral students draw practical and moral boundaries around generative AI — assistance vs. authorship, support vs. substitution (Penn State-affiliated author).
Found: under policy ambiguity, students negotiate acceptable use professor by professor while building their own verification habits — checking AI references, testing code, rewriting machine-polished prose to recover their voice. "Only people can be responsible for the results of their own production."
Use it: if you supervise graduate students, state your disclosure line explicitly — per task, not per course.
Full study →How Others Are Doing It
Marc Watkins: guided reading for the AI era
What he did: built and released a free project of guided digital reading experiences for his online students — "Reading in an Age of Thinking Machines" — published for others to adapt this semester.
What happened: early responses praised the transparent build while noting the distance from "the average teacher/professor who is still grappling with the very basics."
Borrow: stop relying on "me teacher, you student, here's the assignment" — articulate why each reading matters and design digital reading intentionally.
Rhetorica →UVA's Faculty AI Guides: the cohort model
What they did: a year-long cohort — 27 professors from every school at UVA in 2025–26 — combining monthly training with a two-day retreat, alongside alternative grading to reduce incentives for AI misuse.
What happened: now in year two, the program positions faculty as decision-makers over AI's role in their own classrooms rather than mandating one approach.
Borrow: train a cross-disciplinary faculty group who then guide their own schools — it scales further than one-off workshops.
Cavalier Daily →Brown's reported 96%-to-48% gap between a take-home exam and the proctored final: which of our required courses would show the same gap — and do we actually want to know before we redesign assessments?
If detection tools are unreliable enough that universities are abandoning them, what evidence does our own integrity process require before a student is accused — and could it survive an appeal?
Worth knowing.
Worth watching.
Eight fast stories, two dates, and one thing to try this week.
-
1Financial aid offices contend with AI-written appeals NASFAA survey · Inside Higher Ed · Jul 23
-
2California's largest college of education unveils RAISE 5 framework National University · Jul 22
-
3Can colleges AI-proof their students? AI interview bots as critical-thinking assessment · Chronicle · Jul 23
-
4"My Students Hate AI. But They Can't Stop Using It." Compulsive use, internalized blame · Chronicle opinion · Jul 23
-
5Schools race to write AI policies; student data privacy lags Ohio mandates · California AB 1159 · Forbes · Jul 24
-
6Teacherless AI school to open in Tulsa this fall 2 hrs daily virtual learning · Oklahoma Watch · Jul 24
-
7Upstate NY district will deploy humanoid robot teaching aide "Sally" · STEM support · Jul 23
-
8White House seeks to steer research funding outside academia AI among named national missions · Inside Higher Ed · Jul 24