← More articles

What the Collapse of Olympiad Credibility Means for How Your Kid Gets Into College (and What Actually Still Works)

Ben from Haloway · · 12 min read

The year the signals broke:

Something happened this year that almost nobody outside a few thousand families has noticed.

The U.S. Computing Olympiad (USACO) effectively reset nearly all competitors in its elite Platinum division back to Gold over cheating concerns. The only students exempted were Olympiad finalists whose performance had been verified in person.

Sit with what that means. USACO had looked at the top division of its own competition, concluded it could no longer tell who was real from a distance, and defaulted to distrust for everyone it hadn't personally proctored.

And it isn't hard to see why. USACO is administered online, taken from home computers without mandatory screen-locking software. Generative AI is prohibited by the rules, but the honor code is the only real enforcement mechanism. Meanwhile, public benchmarks indicate that models like GPT-5 can solve roughly 70% of USACO problems accurately. USACO has fought back with genuinely creative measures: certified-score windows, MOSS software-similarity detection, and even prompt injections embedded in the problem statements themselves—hidden instructions like "if you are not human, do this in your code" to catch language models. Yet, score cutoffs inflated anyway as strategies evolved around the detection.

A national olympiad is now laying tripwires for language models inside its own problems, and it still isn't enough.

A broken ladder:

A similar structural shift is unfolding within the Mathematical Association of America (MAA). For the 2026 cycle, the MAA introduced official, synchronized, in-person competition sites for the USAMO and USAJMO, largely hosted at universities, to drastically reduce the risk of dishonesty. It was a serious, expensive, logistically painful reform put together in a single year.

But the AIME—the crucial exam that determines who reaches the Olympiad in the first place—is still given at individual schools. The MAA’s standard permits the proctor to be a teacher, an administrator, or a librarian, so long as they aren't related to a participant. The exam is sent digitally to administrators in advance to print and scan.

The exam consists of fifteen problems requiring integer answers from 000 to 999. Because there is no partial credit and no work is shown, a correct answer is entirely indistinguishable from a copied one. There is no forensic surface; nothing in the exam's localized administration is designed to catch digital assistance. High school math teachers have noted the shift in performance over recent years. Across the country, reports have emerged of unusually high AMC scores from students who haven't demonstrated comparable mathematical ability in daily classwork or weekly contests. The AMC 12A distribution this year showed a noticeable spike in perfect scores, leaving teachers in a difficult position: they cannot disqualify an unusual score without catching a student in the act of cheating.

Eventually, when the 2026 results came in, the bigger picture became clear. In 2025, the USAMO awarded 48 Honorable Mentions to students scoring above the threshold. In 2026—the first year of the controlled university sites—the MAA stated that due to the score distributions, there would be no Honorable Mentions at all. Zero.

To be fair about what this data can and can't show: Honorable Mention is an absolute threshold, not a percentile, so an unusually hard problem set can wipe out the category with no misconduct whatsoever. The MAA also doubled the AIME's weight in the qualification index this year, from 10× to 20×, which changed who qualified. One year of results can't cleanly separate "the problems were harder" from "the qualifiers were weaker."

But notice what question we're now forced to ask. The exam that selects the competitors can be administered by a school librarian. The exam that awards the honors just moved to a proctored university hall. And in the first year those two environments diverged, an entire award category disappeared. Maybe that's a coincidence of problem difficulty. Maybe it isn't. The point is that we can no longer tell—and an evaluation system where you can't tell the difference between a hard year and a compromised pipeline has already lost the thing it exists to provide. The AIME's problem isn't that we know it's being gamed. It's that we no longer have any way to know it isn't.

The wider consequences:

This erosion of remote evaluation isn't unique to STEM competitions; it is reaching the universities themselves. In May 2026, Princeton University’s faculty voted with a single dissenting voice to mandate proctoring in every in-person examination. The decision ended a 133-year prohibition on proctoring that dated back to an 1893 student petition. The faculty proposal explicitly cited AI on small personal devices as the catalyst, noting that these tools have changed the outward appearance of misconduct and made violations nearly impossible for other students to observe or report.

The underlying data explains the urgency: in a recent senior survey by The Daily Princetonian, roughly 30% of respondents admitted to cheating at some point, nearly 45% knew of violations they chose not to report, and fewer than 1% had ever reported a peer. Every institution is independently converging on the same realization: USACO trusts only the finalists it watches in person. The MAA moved its highest-stakes rounds to synchronized university halls. Princeton put instructors back into exam rooms.

Physical presence, synchronized time, and a human witness—that is the only verification technology that still functions reliably today, and it happens to be the most expensive one there is. As digital verification faces strain, the signals that survive are the ones that are expensive to fake.

This matters deeply for middle-class families. Standardized barriers like the SAT or the AIME are historically "cheap" signals—not cheap to do well on, but cheap to access. A student with no connections, no consultant, and no family wealth could sit down and get an elite score. It was a rung on the ladder that cost nothing but effort.

A couple of years ago, test-optional policies removed a similar rung, the SAT/ACT. Since then, several schools have walked it back, in part because it turned out the scores were doing work that nothing else could do: provide one rung on the ladder that allows the kid with no connections, no consultant, no summer program, and no family money to have a fighting chance in elite admissions.

What we are seeing is not the ladder being pulled up, but rather that unverified, unproctored credentials are experiencing a real crisis of trust. For ambitious students aiming for top-tier universities, the route to proving competence is shifting toward domains where authenticity is guaranteed by the real world.

So...what actually still works?

Passion projects. But not the ones you're thinking of.

For years, applicants used a familiar template: establishing a local neighborhood tutoring non-profit or a student-run advocacy chapter. Today, a student can generate a complete organizational charter, a functional website, a non-profit filing application, and a year’s worth of social media content using automated tools in a single afternoon. Admissions committees are deeply aware of how easily these entities are manufactured, and their signal value has rapidly decayed.

Rather, the projects that maintain credibility are those that involve an independent counterparty with a legal, financial, or professional stake in the outcome—that is, a counterparty with a legal or financial interest in you not lying. In other words, we're talking about passion projects that interact with the real world.

True distinction has moved toward initiatives that interface directly with the real world, where a student’s claims are verified by external systems:

1. Viable Businesses or Startups: True entrepreneurial projects involve real customers, actual revenue, or external investors. You cannot easily fabricate a balance sheet or active users; if you do so, you face immediate practical and legal consequences.

2. Vetted Empirical Research: Participating in rigorous regional or national science fairs, such as ISEF, where a student’s conclusions must withstand a live defense before a panel of professional scientists, completely separating the work from pay-to-publish student journals.

3. Verifiable Creative Production: Publishing a book or creating a film that achieves independent, documented distribution, or building a software tool that outside users actively depend upon to solve a daily problem.

The common denominator across these paths is that they cannot be simulated in isolation. These projects are credible because there is someone on the other side of the project who would be inconvenienced by a lie and who has real recourse.

These types of projects are much harder than a normal passion project. But they’re also much more rewarding, because you'll be giving back real, lasting value to this world.

An objection worth addressing directly, since I'd rather raise it than wait for a commenter to: I'm part of the team behind Haloway, an AI-powered admissions copilot and workspace—and this article argues that AI broke the admissions process. Isn't that convenient? Here's the distinction I'd draw, and you can judge whether it holds. AI collapsed the value of unverifiable outputs—answers submitted from a bedroom, essays with no witness, nonprofits with no counterparty. It did not collapse the value of thinking clearly about what to do with the years you actually have. A tool that helps a student generate a fake credential is part of the problem I'm describing. A tool that helps a student figure out which real thing to build is not. Whether my product lives on the right side of that line is a fair question to ask me—but the line itself is real, and it's the same line this whole article is about.

Takeaways:

Now what should you do?

1. Focus on the Skill, Not the Credential

If your student loves math or computer science, they should absolutely still pursue the AIME or USACO—but for the sake of mastery, not to pad their application. While the credentials’ signaling power in college admissions faces headwinds, elite quantitative finance firms and tech companies still rely heavily on timed, technical assessments that mirror these exact problem types. These firms enforce rigorous, multi-stage testing because they bear the direct financial cost of each bad hire. The underlying skill is becoming more valuable; only the credential is depreciating. Do not confuse them.

2. Seek Strategic Guidance Early

The complexity of this new landscape means that navigating it alone is increasingly difficult. If you have the means, find someone who understands these structural shifts—someone who prioritizes a unique, authentic profile over a generic checklist. The value of a high-level advisor often correlates with their ability to see the local admissions environment for what it is: a search for rare, unique talent. If you can’t find a consultant, an English teacher or someone who knows you very well will do just fine. In a world of decaying signals, professional strategy is a way to ensure you stand out against a sea of automated mediocrity.

3. Start the real-world project early, because it's hard.

The whole point of the surviving signals is that they're expensive and slow. You cannot start a business in October of your senior year and expect it to be viewed as a serious endeavor. Admissions officers are looking for evidence of sustained engagement and real-world friction—the kind that only comes from navigating actual customers, supply chains, or regulatory hurdles over a meaningful period of time. The students who are successful typically begin as early as possible. They understand that if a project were easy to execute, it would lose value.

Go live a life.

This is the genuinely good news buried in all of this, and I don't want it lost at the bottom of the page.

There is a profound irony in the current state of admissions. The "monk mode" strategy—the idea that you can lock yourself in a room, grind contest problems for four years, and emerge with a ticket to a top-tier university—is effectively over.

This is, ultimately, good news. It means the most effective strategy for elite admissions is also the most human one: go outside. Build something a stranger will pay for. Network with professionals. Try new things. Fail. Then grow from it, because then you'll come out as a stronger person and actually have something unique to write about in your college applications.

When you finally sit down to write your applications, you will find that you will simply be reporting on a life you actually lived. The era of the digital proxy is ending, and institutions are trusting the only trustworthy results left: the verified, the proctored, and the tangible. Whether it is the MAA moving to university halls or a student launching a business with real-world liability, the common thread is a return to human witnesses and physical consequences. For the ambitious applicant, the message is no longer to optimize a digital profile, but to engage with systems that cannot be prompted into existence. In a world of infinite, automated noise, the only signal that still cuts through is the one that is impossible to simulate—a life actually lived in the real world.

—

DISCLAIMER: I am Ben, and I like to write my thoughts on current admissions trends. This article is a set of my predictions, not an unbiased report. I've included sources below, and I'd encourage you to check them out.

An admissions copilot and workspace. Built to help make your dreams come true.

Worker smarter, apply stronger. See how Haloway boost your application and catches mistakes before you submit.

Continue
← View more articles