Resilience or Subordination to AI: Challenges for Chile

In just forty-eight hours, between June 2 and 3 of last week, three events reshaped the changing landscape of emerging technologies. An academic laboratory at the University of Toronto, Canada, demonstrated the anticipated risk that AI models based on open-source code could transform into malware (worms) that are extremely difficult to contain in cyberspace.[i] The White House, reluctant to regulate AI development companies, decided to adopt measures aimed at monitoring the most dangerous cyber capabilities of advanced models and coordinating with the industry to protect critical infrastructure.[ii] Europe, which had previously focused on legislating AI regulation, has decided to implement measures aimed at increasing its technological sovereignty.[iii] It is not that one event caused the others, as both the U.S. order and the European package had been in the works for months. What is relevant is that all three are symptoms of the same structural change. Analyzing them together allows us to infer that the logic of control over artificial intelligence in the cyber domain is transforming, as it shifts from containing a few actors to prioritizing resilience and sovereignty. For a country like Chile, which is predominantly a user of technology, this shift deepens the need to make decisions.

The Open Access that Shakes the Board

A team from the University of Toronto, led by Professor Nicolas Papernot, built an AI-powered “worm” capable of inferring new modes and adapting various attacks according to the target machine or system, which also propagates without human intervention. The novelty of this creation is not the prototype itself, but that its main inputs come from open sources (open source) available on the internet. As Dr. Papernot warns, there would be no single patch to stop something that mutates. However, the study itself acknowledges that there remains a considerable gap between what can be achieved in a laboratory and the damage that can be inflicted in the real world, as AI systems are still partly unpredictable.

On the other hand, this experiment confirms that AI will continue to tilt the playing field between attacker and defender. The offensive capability sufficient to cause harm is already disseminated through open sources, while the defensive side requires access, AI models, talent, and resources that must be financed and sustained over time. Additionally, the attacker needs only one success, while the defender must always be right. This asymmetry goes beyond the costs associated with attacking adversary systems and defending them by potential victims, as both sides of the equation require similar capabilities and resources. The inequality that consolidates is of a different nature, as it operates on the margin of error and the ease of access to offensive capability, rather than on cost.

It could be objected, as the paper by Professor Papernot does[iv], that the “worm” is neutral since it can also be modified to protect systems, and therefore, the power to have such programs depends on what one intends to do with them. But this neutrality is theoretical, as in practice the harmful version is already circulating within reach of anyone who knows how to configure it, while the protective version still needs to be built. In this unequal race, Chile, which is consolidating its cyber capabilities, starts at a disadvantage.

Washington, Regulation, and the Pioneer Company

The notable aspect of the Executive Order signed by President Trump on June 2[v] is not that it regulates, but that the same administration known for dismantling the “bureaucratic obligations” of its predecessor establishes a complex state process. Its main elements include a classified assessment of capabilities by the National Security Agency to designate “covered frontier models,” a center for exchanging and coordinating vulnerabilities led by the Department of the Treasury, and a channel for access to advanced models for critical infrastructure operators. With this, the regulation of AI models shifts from the consumer to the state, as an additional function to its role as the primary responsible party for national security.

In this regard, it could be said that the state arrived second. Weeks earlier, Anthropic had restricted the dissemination of its most advanced model, considering it too dangerous, initially providing it only to a group of organizations to address their vulnerabilities. The Executive Order essentially systematizes as public policy a procedure that a company had already voluntarily undertaken. Just as in other areas of the defense industry, in the realm of AI, the private sector is not only part of the response but has the capabilities to be a pioneer in security.

These events confirm that today states cannot act alone in the face of advancing technologies. The U.S. Executive Order is deliberately multi-agency and involves voluntary collaboration with the industry. A “whole-of-nation” approach, where government, businesses, academia, and infrastructure operators act in coordination. It is a ratification that technical capacity no longer resides in the state but is distributed across an ecosystem. In the case of Chile, it is imperative to articulate public needs and private initiatives that contribute in that direction.

Regulating the Foreign and Subordination on the Horizon

The European Union announced this week its Technological Sovereignty Package, and the data motivating it is eloquent. To date, the EU produces less than 10% of the world’s semiconductors and is almost entirely dependent on the United States and Asia for the advanced chips that power AI. This hardware dependence is compounded by reliance on the most powerful or frontier models, which remain closed and concentrated in a handful of U.S. and Chinese laboratories. The package aims to reverse this situation by promoting domestic semiconductor production, expanding European data center capacity, and introducing sovereignty requirements for cloud and AI providers in sensitive sectors such as banking, energy, and health. Europe does not dominate either the physical input that sustains AI or the cutting-edge capability that defines it, and reducing this dual deficiency is the underlying purpose of the initiative.

However, a power that depends on others for the essential components of its technology can regulate as much as it wants within its borders, but the fundamental decisions, such as who manufactures the chips, who trains the models, and under what conditions they are accessed, remain in the hands of external actors. This means depending on those over whom one has no power, and it is precisely the fate that Europe seeks to avoid by moving from merely regulating to also prioritizing the construction of its own capacity.

Conscious Dependence

The events described may seem distant, characteristic of great powers and enormous budgets, but they actually impact Chile directly. The threat demonstrated at the University of Toronto does not distinguish between the size of a country or its level of development, as once a tool of this type circulates on the internet, it reaches a power just as easily as a medium-sized state. The difference between one country and another is not whether it will be reached, but in its capacity to confront the attack and recover; and in that scenario, Chile would be at a disadvantage.

Chile does not have the advanced developers and laboratories that the United States possesses, nor the resources that Europe seeks to allocate to build sovereignty, and this is to be expected in a country that is, above all, a user of technology and not a producer. Furthermore, the U.S. document mentions types of infrastructure that are particularly exposed, which reflect much of what sustains daily life in our country.

In light of this scenario, there are three decisions that Chile can adopt. The first is to accept that total security does not exist and focus on resilience; this means incorporating defensive artificial intelligence into critical infrastructure (energy, water, telecommunications, health, banking, and transportation) as a state policy, so that the country can detect, contain, and recover from foreseeable future attacks. If offensive capability would be available to any type of actor, defense and recovery are the variables that are within our control. The second is to consciously choose whom we depend on, evaluating technology providers critically rather than assuming an autonomy that we do not currently possess. In this area, national regulation has a limited role, as Chile will not produce the most advanced models required in the future, and thus its efforts would be more effective in setting conditions for access, transparency, and security for those providers. The third is to strive for a minimum degree of sovereignty in those areas where an external service interruption, decided outside the country, would leave Chile without the capacity to react.

Ultimately, the important question is who ultimately controls the technology on which we all depend, and for an open country like ours, this question is especially urgent. Chile still has time to make the decisions that will allow it not to be at the mercy of what others decide for it.

---

[i] New York Times. Scientists Find Way to Supercharge Dangerous Computer ‘Worms’ With A.I. June 2, 2026. https://www.nytimes.com/2026/06/02/technology/scientists-find-way-to-supercharge-dangerous-computer-worms-with-ai.html?campaign_id=9&emc=edit_nn_20260603&instance_id=176596&nl=the-morning&regi_id=92547472&segment_id=220885&user_id=97af3860a63d30d069257fbddbdf8ffa

[ii] The White House. Executive Order: Promoting Advanced Artificial Intelligence Innovation And Security. June 2, 2026. https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/

[iii] European Commission. Commission proposes tech sovereignty package to strengthen Europe’s digital autonomy and resilience. https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1187

[iv] New York Times. Scientists Find Way to Supercharge Dangerous Computer ‘Worms’ With A.I. June 2, 2026.

[v] The White House. Executive Order: Promoting Advanced Artificial Intelligence Innovation And Security. June 2, 2026.