Why Employers No Longer Trust College Degrees: Grade Inflation, the SAT Revival, and the Rise of Hiring Tests

Harvard caps its A’s, fourteen hundred professors want the SAT back, and employers have already built their own tests. Where the college degree’s signal actually went.

Suppose you graduate this spring with a 4.0 from the most famous university in America, and you apply for a job at a software company owned by Vista Equity Partners. Before anyone reads your resume with much interest, you will sit the Criteria Cognitive Aptitude Test: fifty questions, fifteen minutes. You’ll also complete a personality profile. Everyone does. When Vista buys a company, every employee and every applicant takes the test (at the last public count, roughly 125,000 test-takers for 6,000 hires). And if your score comes back looking too good, they may ask you to sit it again, in a room, with someone watching.

A reader, Frank Hecker, pointed me to Vista in the comments on The Sanctioned Hack, and I really haven’t found a better image for where higher education actually stands. A private equity firm looks at a Harvard transcript — the most expensive quality signal American civilization produces — and declines to believe it.

So it built its own credential. Then it looks at its own credential and declines to fully believe that, which is why the retest exists. Somewhere in that hiring workflow is everything this piece is about.

When a credential stops signaling, the signal doesn’t disappear. It migrates.

Right now it’s migrating in two directions at once, and the fight over which direction wins (more precisely: who gets stuck with which) runs underneath a month of higher-ed news that’s mostly been covered as separate items.

My argument: they’re one item.

The strangers stopped believing first

A college degree is two things bolted together. It’s a record of what you did, and it’s a signal to strangers about what you can do. The Sanctioned Hack was about the first bolt coming loose: the record no longer reliably tracks learning, because every incentive in the system pays people to make sure it doesn’t. This piece is about the second bolt. A signal that strangers don’t trust is worth nothing to them, and it was always the strangers (the employer, the grad school, the licensing board) that the whole apparatus was for.

Hollis Robbins, in a piece from late June called For Whom the Bell Curves that I’ve been tossing around ever since (you do just have a way, Anecdotal Value), gets at what the apparatus was doing. A grade curve, she points out, is a cheap proxy for expensive judgment. It replaces the personal note a professor once signed (”I taught this student, I know her work, I’ll put my name behind her!”) with a two-parameter summary that lets busy strangers rank, in her phrase, “each other’s children.” The curve exists so nobody has to actually know anything about you. That was the deal: universities compress human judgment into a number, and the rest of society agrees to treat the number as real.

The deal is not in good shape. Look at who broke it first: not the grade inflaters (though they did a number too), but the strangers.

Employers screening applicants by GPA fell from about 73 percent to the low 40s over the past decade or so, per the National Association of Colleges and Employers. That collapse predates most of the current panic. Meanwhile the number they stopped reading kept inflating: a Berkeley working paper tracking public-university courses through 2025 finds A’s surging specifically in AI-exposed courses: the ones where the work can be machine-generated and the assessment can’t tell. Grades got cheaper to mint at the exact moment their consumers stopped accepting the currency.

If you want the live demonstration, it arrived yesterday. Roberto Serrano, an economist at Brown, gave his welfare economics course a take-home midterm this spring (a compassionate call, after a campus shooting in December), and enrollment nearly tripled on the news. The midterm average came back 96 percent, in a course that historically runs 65 to 80. So he moved the final into the room, proctored. Eighteen students dropped, nine more never showed, and the average fell to 48.6 percent, the lowest in his nearly twenty years teaching the course. The flight wasn’t random, either: of the 27 who wouldn’t sit the proctored final, El País counted 22 with perfect 100s on the take-home. Same course, same semester, same material. The variable that moved was whether anyone was watching. Any political scientist will tell you what happens next, because it’s what always happens when an official channel stops carrying real information: the information finds another channel.

The restorationists

What’s happened instead, this summer, is a restoration movement. Four fights (and counting), same project.

Fourteen hundred University of California faculty, including seven of the nine math department chairs, signed a letter demanding the SAT/ACT math section back for STEM applicants, after a UCSD report found first-years with below-middle-school math skills had grown nearly thirtyfold since testing ended. Harvard’s faculty voted to cap A’s starting in fall 2027 (I wrote about the pre-history of that vote in Harvard Won’t Fail You). The Manhattan Institute proposed that states mandate an “inflation-adjusted GPA” on public-university transcripts: a second number, discounted by each course’s median grade, printed next to the first one like a real price next to a sticker price. And on Tuesday, an economist at Bowdoin suggested the problem is better understood as shrinkflation (the grade holds steady while the rigor inside it shrinks) and proposed publishing grade distributions so employers can do their own discounting.

Put the four fights side by side, though. One camp wants a number it can believe before the student arrives; the other wants a number it can believe after. Robbins caught the kinship first: both camps, she writes, “are annoyed that a college degree no longer signals anything.” And the two camps are usually enemies: one wants a signal money can demonstrably improve (the 2025 College Board data puts the national average at 1029; students from families above $200,000 run roughly 200 points higher), the other wants a signal money supposedly can’t touch. But they’re on the same side of the deeper divide. Both are trying to resurrect the proxy. Both believe the old public signal can be repaired in place, recalibrated, made legible again.

And they’re winning, procedurally. Yale reinstated the SAT in May, Columbia became the last Ivy to fold in June. The UC Regents get a briefing next week. But watch the two clocks. The UC process runs through faculty workgroups convening in October, reports due May 2027, then Regents approval; the earliest a restored requirement touches an applicant is fall 2028, more likely 2029. Vista is testing this afternoon.

My problem with the restoration project isn’t the diagnosis; the restorationists are (very) right about the rot. But you can’t curve your way back to trust.

An inflation-adjusted A attached to a hollow course doesn’t reveal the hollowness; it launders it, putting an official-looking discount rate on an asset that shouldn’t be rated at all. And the milder transparency versions have already been run. Cornell published median grades online starting in the late ‘90s, and enrollment flowed to the easy classes so reliably that inflation accelerated, a result documented with admirable honesty in the Manhattan Institute’s own brief. Sunlight turned out to be a shopping catalog.

The restorationists are polishing a number the people it was built for no longer read. Which raises the question the restoration debate keeps not asking: if the strangers stopped trusting the public signal, what are they trusting instead?

Where the signal went

Follow the trust. It leads somewhere specific. Four somewheres, actually.

Employers are building their own credentials. Vista is the mature version, but the “purest” one is Palantir’s Meritocracy Fellowship: the company now hires high school graduates directly, no degree involved, screened on (of all instruments) an SAT score of 1460 or better, followed by a paid trial period.

Run that back. A company decides the university can no longer be trusted to certify talent, routes around it, and the instrument it reaches for is the SAT: the same standardized test the universities spent five years dropping. The market’s replacement for the college degree is the college entrance exam. Administered privately, for its own purposes, under its own rules. Across the economy the pattern is the same at lower resolution: 81 percent of employers now use skills assessments somewhere in hiring, up from 56 percent in 2022.

Some employers claimed to drop the degree instead. Don’t believe most of them. The “skills-first hiring” wave, where companies announce they’ve removed degree requirements, turns out to be almost entirely announcement: Harvard’s Joseph Fuller and the Burning Glass Institute found the policy change affected fewer than 1 in 700 hires. The press release was the product. (The sincere version of this idea lives on in things like the skills-based federal contracting bill the House passed unanimously in February. One to watch, not yet one to lean on.)

AI assessment is coming for the question itself. Robbins’s hopeful horizon, and I’ll state it fairly because I think it’s the strongest version of optimism available here: the only thing anyone ever wanted the transcript to answer is, in her words, “can this person do this thing?”

AI may soon assess that directly, before college and after, better than any proxy ever did. If the curve was a workaround for the expense of individual judgment, the workaround’s reason for existing is expiring. She may well be right. Hold that thought until the last section.

And the professional-exam model was here all along. Another reader on the Sanctioned Hack, an actuary posting as SamChevre, reminded me that his field has run “no teacher, rigorous credential” for a century. Actuarial exams don’t care where you learned the material, or whether anyone taught you at all. The credential works because the people who build and grade the exams have no stake in the pass rate. That’s the entire trick, and it’s precisely the incentive higher ed inverted: the contingent instructor needs friendly course evaluations to keep her job, the online platform sells credits, the accreditor bills the provider. Nearly everyone certifying learning gets paid more when more people pass. Decouple the grader from the revenue and the rot has nowhere to live. The honest limit: this scales beautifully in bright-line fields (actuarial science, accounting, the bar) and gets expensive fast in fields where honest assessment means a skilled examiner asking you to defend an interpretation out loud. Some competence costs more to certify than a multiple-choice sitting. That may just be the price.

Four destinations, one pattern. The public signal got cheap to fake, so trust migrated to private signals that are expensive to fake: the proctored sitting, the paid trial, the exam with no teacher attached. Nobody planned this. It’s what trust does when a channel goes bad.

Who gets which signal

If the story ended there, you could tell it as markets doing their thing, mildly embarrassing for the universities, ultimately fine. It doesn’t end there, because the migration isn’t distributed evenly. It’s splitting along the same line the whole system is splitting along.

At the top, the degree still signals, so employers respond to credential distrust not by testing Harvard graduates but by trusting a shorter list of schools harder. Recruiting analysts at Veris Insights find the share of firms recruiting from a designated short list of elite universities rose from 17 percent in 2022 to 26 percent now. “Talent is everywhere” is out; the safe thirty schools are back. If your alma mater is on the list, your signal survived: concentrated, actually, into something more valuable than before. (The Elite Mentorship tier keeps its vouching ability, I guess, cough.)

Everyone else’s graduates get the funnel. The 3.8 from a regional public — the institution that took the enrollment hit, adjunctified the faculty, bought the online program in a box, the one being squeezed toward the Credential Factory tier — arrives at the employer pre-discounted, and the employer substitutes its own gate: the CCAT, the personality profile, the AI interview. The gate has no syllabus and no appeals office. There’s already a thriving prep industry for it, which means the money-buys-the-signal problem that made people hate the SAT has been rebuilt intact, one layer down, minus the public scrutiny.

And I’ll grant the tests their best case, because it’s real and personal: for a sharp kid from a school no recruiter visits, a blind cognitive screen can be the fairest shot she gets. Robert Smith pitches Vista’s test as exactly that (a bypass around where you grew up and where you went to school), and sometimes it surely works that way. The trouble sits one level down: when the public test misfired, you could audit it. Nobody outside the firm gets to check this one’s math.

I teach at a public university, so let me say this part plainly: the students being routed into the unaccountable version of the signal are precisely the students the public version was invented for. The kid with no network, no legacy admit, no family friend at the firm — the transcript was supposed to be her instrument for being believed by strangers. The elite kid never needed it. The signal migration takes the one credential that was at least nominally public, auditable, and contestable, and replaces it, for the bottom two-thirds of the map only, with a proprietary score she can’t see, generated by a test she can’t study for honestly, feeding a decision nobody has to explain.

Which is also my answer to the access-first side of the transfer debate. Doug Lederman engaged the Sanctioned Hack generously at IHE yesterday, arguing that in credit transfer the greater risk is exclusion, not gaming. On the intake side he may be right. But lower the gate inside the university and the real gate doesn’t disappear. It moves outside, where nobody owes your students an explanation.

The gate you can’t argue with

Which brings me to the part of this that’s really bugged me since I started working on this, because it’s happening in the same news cycle and nobody’s connecting the two.

For fifty years, much of the reason American employers didn’t run their own testing gauntlets was a public one: Griggs v. Duke Power (1971) and everything built on it, which required employment tests with discriminatory effects to prove they measured something the job actually required. You could fight a test. There was a body to complain to and a doctrine to complain under. That’s the machinery that pushed employers toward the degree requirement in the first place: the transcript was the screen you couldn’t get sued over.

Well, that machinery was dismantled last month. The EEOC’s new enforcement plan walked away from disparate-impact enforcement on June 4; five days later the DOJ declared the EEOC’s disparate-impact guidelines unconstitutional outright. The federal accountability layer for private employment testing effectively shut down in the same month the private tests became the gate.

What’s left is a patchwork, not a floor. A federal judge in California ruled in June that state law reaches Workday’s AI screening tools wherever they’re used, because the tools are built there. That’s a real channel, and with 80-plus percent of large employers running AI screens, not a small one. But your recourse now depends on which state’s courts can reach which vendor. The kid sitting the personality profile in Ohio is not similarly situated to the one in California, and neither can see the rubric.

So what did the market’s solution actually reinstate? Universities spent decades outsourcing the verification of competence to a piece of paper: no room, no examiner, no watching anyone do anything. The private replacement brings back the room. The proctor. The live demonstration under observation. Employers are rebuilding, at their own expense, exactly the in-the-room assessment the universities abandoned, which tells you the demand for real verification never went anywhere. It just stopped being met publicly. Serrano’s proctored final and Vista’s proctored retest are the same room.

Robbins is right that “the day of the proxy may be over,” and she’s right that what could replace it — direct assessment of whether this person can do this thing — would be better than any curve ever was.

But here’s where I get off her train: better is not the only axis. The old signal was decaying, gameable, and increasingly hollow, but it was public. You could FOIA it, sue it, petition it, cap its A’s, write angry op-eds about it, and — as fourteen hundred UC professors just demonstrated — organize to change it. The new signal is being assembled right now, by private equity firms and assessment vendors and AI companies, and it may well measure competence more honestly than the transcript ever did. There is no mechanism by which you will ever get to argue with it.

The restorationists are fighting to fix the gate everyone can see, on a timeline that runs to 2029. The replacement gates are already open, already sorting, already sorted by class. And the law that once made private gates explain themselves went dark last month.

When the signal migrated, the accountability didn’t follow it. I’d like to tell you where that ends. What I can tell you is who it ends on, and that part of the story is going to be… less than uplifting.

---

Sacred Cow BBQ covers higher ed, politics, institutions, and the trust questions underneath. Two of this piece’s anchors — Vista and the actuarial exams — came out of the comments on The Sanctioned Hack, which is to say: arguing with me works, and the comments are open. The university map is the tool behind the bifurcation claims here — 1,556 four-year institutions, four quadrants, data open. Find your dot. And if you’re on the hiring side of this — if your firm runs a CCAT, an AI screen, or a short list of trusted schools — I’d genuinely like to hear how the sausage gets made.

---

The paper trail

The restoration fights: The UC faculty letter (Inside Higher Ed) and the fuller EdSource report · Harvard’s A-cap vote and the Crimson’s decade of grade data · The Manhattan Institute’s inflation-adjusted GPA brief (Arnold) · The shrinkflation argument (Khan, IHE) · The SAT dominoes (Forbes) · For the other side: grade inflation as moral panic (Bronzini-Vender, Washington Monthly) and the access-first case on credit transfer (Lederman, IHE)

The migration: Hollis Robbins, “For Whom the Bell Curves” — the proxy-collapse argument this piece builds on · Vista’s testing regime (Forbes) · Palantir’s Meritocracy Fellowship (Fortune) · The skills-assessment boom (TestGorilla) · Why skills-first hiring was mostly a press release (Fuller/Burning Glass) · AI and grade inflation (Berkeley CSHE) · The Brown take-home experiment (Whitford, IHE) and the El País account with the fuller numbers · Why the future of college could look like OnlyFans (Kang, New Yorker — the Robbins profile)

The accountability story: DOJ declares the EEOC’s disparate-impact guidelines unconstitutional · The EEOC’s retreat from disparate-impact enforcement · Mobley v. Workday, the June ruling (HR Dive) · The elite short-list retreat (Fortune/Veris)

Earlier in this series: The Sanctioned Hack (the record) · Harvard Won’t Fail You (the grades) · The Bifurcation of Higher Education (the map) · The Collapsing Value Proposition (the bundle)