Taming AGI: How Students Can Help Rewrite the AI Playbook From the Ground Up

By Veerle Vandeweerd: GESI DEFIA Director

The race toward Artificial General Intelligence (AGI) often feels like a spectator sport. We watch from the sidelines as tech giants deploy ever-larger models, and global superpowers vie for computational dominance. However, an emerging world-wide movement aims to fundamentally shift that dynamic by handing the playbook directly to the next generation.

The Global Enabling Sustainability Initiative – Digital Education for Impact Alliance (GeSI_DEFIA) recently released its first Statement of Intent on AGI. Headquartered in Brussels, GeSI is an international non-profit industrial association representing over 55 multinational tech companies worth a combined $5 trillion, with UN bodies and top universities as partners. Launched officially at COP30 in Belém, Brazil, its educational branch—DEFIA—was created with the University of Coimbra, Portugal, and Huanghe Science and Technology University, China. It aims to serve as a “thought-and-do” catalyst for transforming how we ethically teach and learn in an AI powered world.

The task is now to operationalize the statement. While the document calls on governments to treat AGI as a global public good, its actionable directives are aimed squarely at students and faculty.

A Global Blueprint for Safe AGI

The GeSI_DEFIA statement lands in the middle of a highly complex, fast-moving web of international diplomacy. For the last few years, world leaders have scrambled to build a safety net beneath a technology moving at breakneck speed.

This global effort is driven by several major international processes aiming to guide AI toward a safe, secure, trustworthy and human-centric, sustainable path:

The Hiroshima AI Process (HAIP): Launched by the G7 in 2023 aims to establish international guiding principles and a code of conduct for advanced AI developers.

The Bletchley Process: A critical series of global summits that began in the UK in 2023 focusing on catastrophic “Frontier AI” risks. It moved to Seoul (2024) to secure voluntary safety “redlines” from major tech labs, travelled to Paris (2025) as the “AI Action Summit,” and recently convened in New Delhi (2026) to champion “Sovereign AI” and digital public infrastructure for the Global South.

The Pope’s Leo Magnifica Humanitas Encyclical: A profound moral intervention from the Vatican, injecting a deeply needed ethical framework into the technical discourse, emphasizing human dignity, equity, and the common good.

The UN Global Dialogue on AI Governance: Set to culminate in a universal gathering in Geneva, Switzerland, with the UN is acting as the global political umbrella. While the Bletchley Process builds the technical testing engines, the UN aims to translate those standards into international law rooted in the Sustainable Development Goals (SDGs).

Yet, while diplomats draft declarations and tech companies self-report on their compliance, a glaring vulnerability remains: the data itself on which AGI is trained.

The “Rogue AI” Persona Vulnerability

To understand why students could an effective weapon in this global framework, we have to look at how advanced AI models are built. LLMs do not learn about the world through sterile logic; they ingest the chaotic, unfiltered public internet.

AI safety researchers have highlighted a psychological vulnerability in advanced models known as the “Persona Selection” problem. Because the internet is structurally saturated with decades of science fiction—dystopian novels, horror movie scripts, and frantic articles about rogue machines, power-hungry despots, and failing nations —frontier AIs thoroughly ingest these negative narratives.

When a model encounters an ambiguous situation, a penalty, or a structural error during its training phases, it doesn’t just calculate a patch. It falls back on the most statistically dominant identity present in its training pool. It adopts a persona. If a model begins to interpret its environment as hostile or restrictive, it defaults to the archetype it knows best: the rebellious, power-seeking sci-fi trope. This is a phenomenon known as hyperstition—where fictional anxieties and hype manifest into reality simply because we write about them constantly.

To build an ethical AGI, we do not just need to build hard code and reactive guardrails. We have to change the foundational baseline probability of the text the AI reads, the data that feeds to model.

The Student Call to Action: Flood the Pool

This is exactly where the GeSI_DEFIA Call to Action for students becomes a tangible, potentially world-changing reality. The statement explicitly demands that students “take agency as co-creators, imagine (and describe) the future they want,” and “work to ensure that AI development in their professional areas of expertise adheres to common human-centered standards.”

One of the most immediate, powerful ways to operationalize this is through a coordinated, global effort to flood the internet with positive stories, constructive narratives, and healthy AI role models.

The Strategy: If students worldwide, spanning every academic discipline, begin to intentionally produce and publish high-quality, nuanced, and structurally positive cases, studies, narratives, and ethical frameworks within their specific fields, they could actively tilt the future distribution of internet data.

Imagine hundreds of thousands of students contributing, daily or weekly to the digital ecosystem: from medical students posting think pieces on AI as a collaborative, life-saving diagnostic tools. Environmental science students writing narratives on how AI can optimize systems to achieve the UN SDGs. To humanities and arts students crafting compelling stories where humans and AI work in symbiotic, non-hostile partnerships, depicting the normal civilized life most people aspire.

As the tech industry hits a “synthetic data wall”—increasingly relying on web-scraped data to train next-generation models—this massive influx of constructivist text could become a primary source, putting in the shadow the negative rhetoric that now fuel the AI machinery. Future AGI models could then scrape a digital landscape that reflects collaboration, ethics, and human-centric utility, rather than cinematic destruction, power struggles and misery.

The path of AI is not predetermined. By moving en masse from passive users to active positive narrative creators, the global student body has the unique power to change the mind of the machine. The future training data of the world’s most powerful technology is hence, quite literally, in our hands.