> ## Content Index
> Fetch the complete content index at: https://newsletters.qs.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Why are Latin American students so drawn to entrepreneurship despite the risks?
- URL: https://newsletters.qs.com/why-are-latin-american-students-so-drawn-to-entrepreneurship-despite-the-risks/
- Published: 2026-09-02T06:39:31.000Z
- Updated: 2026-09-02T06:39:31.000Z
- Description: QS Midweek Brief - September 2, 2026. Is it nature or nurture for Latin American entrepreneurial spirit? And what does the gen AI watermark actually mean for academic integrity?
- Author: Anton John Crace
- Tags: QS Mid Week Brief

Welcome! This week, we continue out coverage of Latin America in the lead up to the QS Higher Ed Summit: Americas later this month ([tickets still available](https://www.qs.com/conferences/americas?ref=newsletters.qs.com)). You may be surprised to learn that despite its risks, students in Latin American can be more than twice as likely to be an entrepreneur that the global average. We explore why.

Looking at AI headlines, you may have seen that Anthropic has implemented a digital watermark. We unpack what that means and why it’s happening.

Stay insightful,  
Anton John Crace  
Editor in Chief, QS Insights Magazine  
QS Quacquarelli Symonds

![](https://storage.ghost.io/c/f1/d4/f1d411df-6f49-4c75-a302-f8c9e9f59495/content/images/2026/09/1.aurak.3e182fc01ef8.webp)

---

## **Why are Latin American students more likely to become entrepreneurs**

*By Chloë Lane*

![](https://storage.ghost.io/c/f1/d4/f1d411df-6f49-4c75-a302-f8c9e9f59495/content/images/2026/09/Why-are-Latin-American-students-more-likely-to-become-entrepreneurs-Foleon.jpg)

### **In brief**

- Cultural resilience and job scarcity drive record-high entrepreneurship rates among Latin American university students.
- Beyond necessity, family traditions and lower opportunity costs make self-employment a realistic and respected career path.
- Universities should prioritise practical skills and interdisciplinary networking over theoretical courses to foster successful, responsible leadership.

Entrepreneurship is a risk-heavy career path to choose.

It comes with significant uncertainty and a much higher probability of failure than other careers. Yet, more than one in five (28%) graduates say they plan to become an entrepreneur within five years of graduation, according to the most recent [GUESSS Global Report](https://www.guesssurvey.org/resources/PDF%5FInterReports/GUESSS%5F2025%5FGlobal%5FReport.pdf?ref=newsletters.qs.com), which looks at student entrepreneurship around the world.

In some areas in Latin America, however, this figure rises significantly — 66% in Mexico and 37.4% in Brazil. Countries such as Brazil, Colombia, Mexico and Chile are among those with above-average shares of students who plan to start their own business, both directly after studies and five years down the line.

Despite this, entrepreneurial support at Latin American universities varies. Some universities offer strong support, but many students were found to have not attended even a single entrepreneurship course. Student entrepreneurs also often struggle with access to funding, fear of failure and balancing their studies with their businesses.

Yet entrepreneurship remains a popular path. “We are seeing a clear increase in the proportion of \[students\] who want to develop the skills and knowledge needed to become entrepreneurs,” says Professor Fausto García, Director of the Entrepreneurship Center at IAE Business School, Universidad Austral, in Argentina.

“Particularly in programmes with younger cohorts.”

### Responding to uncertain economic environments

Students at Latin American universities are drawn to entrepreneurship because they have a desire to create a positive difference in the world, notes Professor García.

The technological revolution has made this more possible now than ever before. “It’s creating new possibilities and inspiring a new generation of entrepreneurs who can build companies from the ground up,” he says.

However, Latin America’s young people also often turn to startups out of necessity: more than 80%of new entrepreneurs in Mexico, Guatemala, Argentina and Costa Rica say they chose that career path because jobs are scarce, according to the most recent Global Entrepreneurship Monitor report.

“Beyond personal motivations, entrepreneurs in the region often develop the ability to identify opportunities within uncertain economic environments and changing labour markets,” he says. “This capacity to perceive opportunities amid uncertainty is an important characteristic of the Latin American entrepreneurial mindset.”

Entrepreneurial activity in Latin America has historically been linked to economic cycles, with entrepreneurial ventures increasing during periods of crisis and becoming more moderate in times of greater stability.

In response to dealing with their changing economic and political environment, entrepreneurs in the region are often able to adapt quickly to highly uncertain situations — more so than in the West — giving them the skills every good entrepreneur needs to build a successful business.

“It is one of the distinctive strengths that other entrepreneurial ecosystems can learn from.”

### Entrepreneurship as a natural choice

While economic volatility and political change in the region certainly play a role in the drive for entrepreneurship, they are by no means the only factors.

Instead, it is more of a combination of economic realities, cultural values and personal mindset, says Dr Asghar Afshar Jahanshahi, Professor of Strategy, Entrepreneurship and Innovation at IPADE Business School in Mexico.

Rather than being something students are forced to do, entrepreneurship is deeply rooted in Latin American society. Many students grow up in families that own small or medium-sized businesses or know entrepreneurs within their close social circles.

“As a result, starting a business feels like a realistic and respected career option rather than an exceptional one.”

Family support often provides not only encouragement but also practical advice, networks, and, in some cases, resources while they’re starting out. It makes entrepreneurship a lot more accessible compared to other regions.

### The opportunity cost of entrepreneurship

The opportunity cost of starting a business may also play a role, says Jorge Vincio Murillo Rojas, Assistant Professor of Entrepreneurship at INCAE Business School in Costa Rica. In Latin America, the opportunity cost is often significantly lower than in other regions.

This is because, for many graduates, traditional corporate careers in the region may not offer rapid professional growth or the level of impact they aspire to achieve. “Creating a venture becomes an attractive alternative to shape their own leadership path and accelerate their personal and professional development,” says Professor Rojas.

This is something Dr Jahanshahi also notes, reflecting on his career teaching students at leading universities across Chile, Peru and Mexico, where students often view entrepreneurship as a way to determine their own future, particularly when traditional careers don’t offer the opportunities they’re seeking.

“Latin America's entrepreneurial spirit reflects more than economic necessity. It also reflects a culture of resilience, adaptability and optimism that encourages young people to create opportunities rather than wait for them,” he says.

### What role do universities play in inspiring entrepreneurship?

Universities across the region incorporate entrepreneurship into their degrees in different ways. Some, like INCAE Business School, run and take part in entrepreneurship competitions.

Here, students learn to identify opportunities, test ideas, collaborate with entrepreneurs and investors, and develop the confidence to turn ambitious ideas into sustainable ventures with regional impact.

Andrea García, whose team with three other students won first place in the international PRME Students Innovation Studio, says that business school gave her the contacts and experience needed to create the business. “INCAE pushed us out of that classroom. They connected us to the PRME Innovation Studio with the UN Global Compact,” she says.

“Suddenly we had real deadlines, real judges, real feedback from people with no reason to be kind to us. That changes everything.”

Other schools, such as Argentina’s IAE Business School, are equipped with a dedicated Entrepreneurship Center, helping students through the early stages of their ventures - from the initial motivation to start a company through to its early development.

Student entrepreneurs at IAE are inspired and supported at all stages of their company’s development, with access to interdisciplinary knowledge, networks and research. The school continues to look into how it can better help student entrepreneurs and understand the factors that contribute to entrepreneurial success and failure.

Entrepreneurial individuals are often highly motivated and looking to expand their professional networks. Business school gives them the opportunity to do this through teamwork, case discussions, group projects and interactions with classmates from diverse industries and professional backgrounds.

Schools like IPADE ensure there are always networking opportunities available to students, both inside and outside the classroom. Dr Jahanshahi believes that to better support students, universities and business schools should create more opportunities to bring together students of different disciplines. “Some of the most innovative entrepreneurial ideas emerge when different backgrounds work together,” he says.

There is also an emphasis on responsible leadership - helping students build companies that, while successful, also generate long-term economic and social value, something Dr Jahanshahi believes business schools should actively promote.

“Universities should encourage students not only to become entrepreneurs but also to become responsible leaders capable of building sustainable businesses that positively impact their communities,” he says.

[Is there a right way to teach entrepreneurship? Find out in the full article online.](https://eu1.hubs.ly/H0x%5Ffsb0?ref=newsletters.qs.com)

*Chloë Lane is a gold-standard NCTJ-trained journalist specialising in higher education. A former Content Editor for QS, Chloë has a wide range of experience writing articles for a variety of B2B and B2C publications about topics related to business schools, universities, careers and academic research.*

[![](https://storage.ghost.io/c/f1/d4/f1d411df-6f49-4c75-a302-f8c9e9f59495/content/images/2026/09/2-taif_university.718c215488c9.webp)](https://eu1.hubs.ly/H0x%5FfrY0?ref=newsletters.qs.com)

---

## **Mark my words**

*Dr Shadi Hijazi, Associate Director, Consulting, QS Quacquarelli Symonds*

![](https://storage.ghost.io/c/f1/d4/f1d411df-6f49-4c75-a302-f8c9e9f59495/content/images/2026/09/Mark-my-words-preview.jpg)

Anthropic, the artificial-intelligence powerhouse behind Claude, declared in mid-August that every new version of its AI Assistant would start watermarking the output it produces. Not with metadata attached to the output file, nor with some invisible characters between the letters. The mark would be hidden in the choice of the words in text output, so it would stay with the text when it is copied and pasted.

The consequences are stranger than they appear. If someone wrote a thousand-word report entirely by themself and asked Claude to tidy the grammar and the style, what comes back would, from now on, carry a machine-readable signal saying that AI produced it. A detected mark means the text was "processed by Claude". Proofread, translated, summarised or written from scratch: same stamp. The mark records that a machine passed through the prose. It says nothing about who was driving.

The gap between what the technology proves and what people will assume it proves is going to matter. The watermarks are genuinely clever; but they are also fragile and cannot really help universities.

### **From essay mills to Caktus**

Long before ChatGPT, students who wished to outsource their assignments could hire a human to do it. Paying essay mills and freelance ghostwriters grew into an industry large enough that several countries legislated against it. It was expensive, slow and risky.

Generative AI collapsed the price of that transaction to null. Within months of ChatGPT's launch, purpose-built services stopped pretending to be general-purpose assistants. Caktus AI, one of the earliest among students, marketed itself directly as a study companion that would write original essays with citations. The ghostwriter had become a service subscription.

Universities responded by getting detectors: Turnitin's checker, GPTZero, Pangram and others. They are expert neural networks trained on millions of human and machine samples to judge a text's style. Students responded by using "humanisers", such as Undetectable.ai, QuillBot and StealthGPT, which take AI text output and roughen it until it scores as human. The humaniser industry exists because the detector industry does; and each retrains against the other.

The detectors' record does not inspire confidence. OpenAI built its own classifier in 2023; it falsely accused human writing and was withdrawn within months. A Stanford study the same year ran TOEFL essays by non-native English speakers through several commercial detectors: on average, two-thirds were falsely flagged as machine-made, and almost all were flagged by at least one detector, even as the same tools correctly passed through more than 90% of essays by native-speaking American eighth-graders.

The classifiers were mistaking the constrained vocabulary and formal, predictable prose of second-language writers for the statistical smoothness of a machine; they were measuring linguistic range. Today's tools are better, but the failure mode has merely migrated.

Then the machinery turned on the academy itself. NeurIPS, one of the field's flagship conferences, ran a Position Paper Track in 2026 with a simple rule: AI for copy-editing only. All submissions to NeurIPS’s 2026 Position Paper Track were screened with Pangram, and 178 — 18.4% of the track — were desk-rejected for suspected substantial AI use before ordinary peer review, reportedly without a standard appeal route.

One rejected author then ran several papers associated with the track chairs through the same detector, obtaining AI-likelihood scores ranging from 24% to 69%. Those results were anecdotal and do not establish that the papers were AI-generated, but they highlighted concerns about false positives and the detector’s calibration.

### **Rigging the dice**

Watermarking properly is a cleverer business, and understanding it starts from one fact about how a language model writes. At each step the model does not simply take the obvious next word; it computes a list of plausible candidates with odds attached, then rolls a weighted die. That sampling step is a place where a watermark can live.

The original design, published in 2023 by John Kirchenbauer and colleagues at the University of Maryland, divides the model’s token vocabulary at each generation step into a secret “green list” and “red list”. The division is derived from a key and preceding context, and the model is nudged towards green-list tokens.

A detector with the key can then test whether a passage contains an unusually high proportion of green tokens; unwatermarked text should remain near the expected baseline. Google DeepMind’s SynthID-Text, published in Nature in October 2024 and deployed in Gemini, uses a different refinement known as tournament sampling.

Candidate tokens compete through keyed, context-dependent pseudorandom comparisons, creating a statistical signal without permanently favouring particular words. In a comparison involving nearly 20 million Gemini responses, Google reported no statistically significant difference in user feedback between watermarked and unwatermarked outputs.

Meta has also investigated watermarking and provenance methods suitable for more difficult open-weight settings, where a provider cannot enforce a watermark on a model after its weights have been downloaded. Anthropic has described Claude’s marking system as based on a version of SynthID-Text, although the public information does not establish that it is Google’s exact implementation. In any event, a detected mark would indicate that Claude may have processed the text; it would not prove that Claude originated the writing, particularly where the system was used only for copy-editing.

Code and other highly constrained outputs offer fewer safe alternatives, so watermark insertion and detection can be less reliable; even small edits may remove the signal or alter programme behaviour.

### **Dilute, wash, spoof**

Which brings us to how the marks fail. The crude schemes fail first: invisible Unicode or look-alike characters can often be removed by normalisation or find-and-replace. File-level provenance is a separate layer. C2PA records, EXIF and XMP metadata, along with document properties, can travel alongside text in supported file formats and can often be inspected or stripped.

Open-source utilities are already circulating on GitHub that combine these operations. They claim to remove invisible characters and provenance metadata across formats — including images, PDFs, Word files, HTML and Markdown — while also applying rewriting or paraphrasing intended to weaken statistical text-watermarks. Some advertise compatibility with Claude-, Gemini- and OpenAI-related signals, but those claims are not all independently verified.

Over a weekend, I tried to put this to the test. Using a modern AI Assistant, no handwritten code and, less than half an hour, I created utilities for Android, iOS, macOS and Windows that removed the invisible traces from any text. Such demonstrations show that these layers can be comparatively easy to scrub, but they do not establish that every hidden mark can be removed in every format.

The keyed statistical watermark is different. It is not stored in the file as a detachable object; it is expressed through the model’s sequence of word choices. To weaken the signal, one must dilute that sequence through editing, rewriting, translation, truncation or other transformations. The limitation is therefore not merely that a watermark-removal tool lacks a button for the job: without the provider’s detector, successful removal cannot be independently certified.

The statistical kind can fail through manual dilution. Because the expected pattern depends on preceding context, edits alter the evidence associated with later tokens, while a substantial rewrite replaces much of the original marked sequence. Anthropic says light editing may not remove the watermark completely, whereas a complete rewrite replacing every word will; it also emphasises that the mark may indicate Claude’s involvement in editing rather than authorship of the underlying ideas.

The second, more systemic failure mode is washing: routing marked text through a second language model. A preprint published in July tested reference implementations of three schemes against criteria associated with the standards American courts use to assess scientific evidence. It reported that paraphrasing removed the tested SynthID-Text signal in 58 of 59 cases and removed the marks from both schemes in every valid case.

In practical terms, the authors argued that the tested watermark systems fell short on at least three of the five criteria American courts consider when assessing whether scientific or technical evidence is sufficiently reliable (Daubert factors). That result should not be treated as an assessment of Google’s production system, as the study used a reference implementation of SynthID-Text, and its findings may not transfer to every configuration. It nevertheless demonstrates that paraphrasing is a viable class of attack against statistical text-watermarks, and the courtroom framing should give any academic-misconduct panel pause.

The identity of the second model matters. If Claude paraphrases Gemini’s output, the original Gemini signal may be weakened or lost, while Claude may add its own watermark to the replacement tokens. A local open-weight model, such as Meta’s Llama, can, in principle, wash out the first signal without adding a new provider watermark, provided that no watermarking layer is enabled. The user controls the sampling process and does not have to send the text through a provider’s hosted generation service. That may defeat a particular watermark detector, but it does not make the resulting prose universally untraceable: account records, drafts, metadata and stylistic evidence may still exist.

Commercial “humanisers” can have a similar incidental effect. Tools designed to evade stylistic AI detectors often regenerate or substantially paraphrase text, thereby disrupting token-level statistical patterns. This creates a situation in which detector developers adapt to rewriting tools and rewriting tools adapt in turn. The existence of that contest does not prove that every humaniser defeats every watermark, but it illustrates the fragility of relying on a single signal.

The third failure mode is spoofing, the mirror image of removal and arguably the more serious risk. An adversary who learns enough about a scheme’s statistical behaviour may be able to bias human-authored text towards the pattern preferred by a detector, potentially framing an innocent writer or falsely attributing text to a rival model. This remains a threat model rather than a universal practical capability: the feasibility depends on the secrecy of the key, access to the detector, the scheme’s design and the amount of text available.

Keeping the production key secret can reduce straightforward spoofing, but it also concentrates verification power in the provider. The public may then depend on a proprietary detector or interface to determine whether a signal is present. The EU’s implementation framework recognises that text detection is particularly difficult, especially for short passages, and provides for restricted access to some detection solutions during the initial phase.

[Find out why the mark is legally required in the online article](https://eu1.hubs.ly/H0x%5FfmQ0?ref=newsletters.qs.com)

*Dr Shadi Hijazi is Associate Director of Consulting at QS Quacquarelli Symonds. He advises universities, ministries and higher-education leaders on benchmarking, reputation and strategy, translating complex data into insights institutions can act on, with a growing focus on the responsible use of analytics and AI.*

[![](https://storage.ghost.io/c/f1/d4/f1d411df-6f49-4c75-a302-f8c9e9f59495/content/images/2026/09/3.-Ajman-University-2.webp)](https://eu1.hubs.ly/H0x%5FfvB0?ref=newsletters.qs.com)

t