A Meta–Chilean police agreement to detect threats against Kast exposes the legal vacuum around surveillance

By Diario Regionalista Antofagasta · September 22, 2026.

Chile recently saw the arrest of people who posted alleged threats on social media against President José Antonio Kast and other officials; the source does not specify how many cases there were or which agency carried out the arrests. The cases, as this outlet reported, originated in reports associated with Meta Platforms Inc., the company that owns Facebook, Instagram, WhatsApp, Messenger and Threads, and which maintains cooperation mechanisms with police forces to respond to content that may constitute a crime or pose a security risk.

The episode has reopened a debate in Chile that goes beyond these specific cases: what limits apply to the intersection of content moderation, artificial intelligence and criminal prosecution. Sebastián Castillo, president of the Colegio de Periodistas de Antofagasta, put it in these terms: the same detection infrastructure that makes it possible to identify a real threat against a public official can also, absent clear criteria, be used to monitor legitimate political criticism.

Castillo does not dispute that threats should be investigated. His argument is a different one: in Chile "there are still no regulatory frameworks that can produce guidelines allowing proper oversight of how artificial intelligence is used," even though —as he explained— legislation on the matter is currently in progress. Without those guidelines, he argues, the decision about which content warrants police intervention rests on criteria that today cannot be publicly audited.

The association's leader went further, claiming that the authorities "of the day" are using these capabilities more to "repress those who speak out against the government" than to solve outstanding crimes. It is a serious political accusation —against the current Chilean government's use of these tools— that is best read for what it is: the public position of a professional association, not an independently verified fact. Castillo himself frames it as a warning and not as a legal complaint: he calls for "judgment, specialization, professionalism and ethics," not for the agreement to be suspended.

As background, Castillo cited the Índice de Ciudadanía Digital from Fundación País Digital, a study that assessed knowledge, usage, best practices, civic participation and cybersecurity: 54% of respondents did not meet the set of basic digital skills evaluated. The gap, he noted, cuts across society and is not explained by age alone, but also by socioeconomic and migration factors and by access to new technologies. The figure matters because it describes the real asymmetry of the problem: a large share of the public is in no position to understand, let alone audit, how an automated system decides that one of their posts constitutes a threat.

In general, this type of agreement between large platforms and police forces is neither a Chilean peculiarity nor an invention of this government: Meta, like the other major networks, has for years operated formal channels for responding to legal and security requests, supported by automatic classification systems that prioritize which reports reach human review and how urgently. What is distinctive about the Chilean case is not the tool, but the absence of an AI law defining who audits those systems, by what criteria and under what supervision, when they are applied to a function as sensitive as the safety of a head of state.

Our reading is that this episode works as a good barometer of a problem we will see repeated in more countries as governments lean on platforms' detection infrastructure for public security tasks. The short-term risk does not lie in the technology itself —detecting a real threat against a public official is a legitimate and even desirable use of these capabilities— but in the lack of a framework that clearly separates prosecuting a crime from surveilling dissent. Without that clear line, the same system serves to protect any government or to silence its critics, and the difference is left to whoever holds power at any given moment, rather than to a stable rule.

The answer we defend at Zendoric in any AI governance debate is neither to reject these tools nor to give them a blank check: it is the one Castillo himself demands, subjecting them to public criteria, specialized oversight and independent auditing, something Chile —like most countries in the region— still lacks today. In the long run, well-governed AI is precisely what can make it possible to distinguish a real threat from legitimate criticism with greater precision, freeing up police resources to investigate serious crimes instead of monitoring public debate. But that outcome does not arrive on its own: it depends on regulation getting ahead of practice, before today's exception becomes tomorrow's norm.