AI Regulation Tracker  /  Legislation pending

Chile AI systems bill clears the lower house and moves to the Senate with EU-style risk tiers

Chile's Chamber of Deputies approved boletín 16821-19, which would sort AI systems into four risk levels and impose proportional duties on developers and deployers. It is a pending bill in the Senate, not yet law.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

Chile has taken a concrete step toward a general AI law. On October 13, 2025, the Chamber of Deputies approved boletín 16821-19, titled "Regula los sistemas de inteligencia artificial," in particular and sent it to the Senate for its second constitutional stage. The measure, submitted by the executive and coordinated by the Ministry of Science, Technology, Knowledge and Innovation, was merged with an earlier initiative, boletín 15869-19. It is now under study in the Senate's Commission on Future Challenges, Science, Technology and Innovation, with the Finance Commission also involved. This is a bill in progress, not an enacted law, and no effective date has been set.

A four-tier, risk-based structure

The bill sorts AI systems into four risk categories: unacceptable, high risk, limited risk, and no evident risk. The design borrows openly from the European Union AI Act and from ethical principles set out by UNESCO. The core idea is proportionality. Obligations rise with the level of risk a system poses to people, rather than applying uniformly to every use of the technology. Systems in the unacceptable category would be prohibited outright. Systems classed as high risk would carry the most detailed compliance duties. Lower-risk systems would face lighter requirements, weighted toward transparency.

What high-risk systems would have to do

For high-risk systems, the bill would require a risk management system built as a continuous and iterative process spanning the full lifecycle of the system, subject to periodic review so that failures, malfunctions, and deviations from the intended purpose are detected and reduced. It also frames a set of principles that operators would need to observe, including human oversight, technical safety, data protection, transparency, accountability, and protection of consumer rights. In practical terms, a firm deploying a high-risk system would be expected to show its work: how the system was tested, what risks it presents, and how those risks are controlled over time. The bill draws a line between the party that develops a system and the party that puts it into use, and it assigns duties to both roles. That distinction matters for supply chains, because a Chilean deployer that buys a high-risk model from a foreign vendor would still carry its own compliance responsibilities under the proposed text.

Prohibited practices

The proposal lists uses that would be banned outright. These include subliminal manipulation techniques that induce harmful behavior or impair a person's decision-making, with narrow carve-outs described for therapeutic uses carried out with explicit consent. The prohibited-practices tier is the sharpest edge of the framework, because it removes certain applications from the market entirely rather than conditioning them on compliance.

Oversight and penalties

The bill contemplates a technical advisory council on AI to support the Ministry of Science, including proposing which systems should be treated as high risk and reviewing implementation over time. On enforcement, the text sets a penalty range reported from 5,000 to 20,000 UTM depending on the severity of the infringement. UTM is a Chilean inflation-indexed unit used to set fines, so the peso value would track that index at the time of any sanction. Readers should treat these figures as the current bill language, which the Senate can still amend.

What it does not do yet

The most important limit is procedural. Because the bill sits in the Senate's second constitutional stage, its provisions do not bind anyone today, and the Senate can change the tiers, the obligations, the council, and the penalties before any final vote. The Commission on Future Challenges has continued to gather input, including a seminar titled "Hacia una regulación de la inteligencia artificial en Chile," reported as preliminarily scheduled for April 24, 2026, with academics, industry, and public-sector participants. That process signals active work, not a settled outcome.

For a US developer or deployer, the cross-border read is straightforward. Chile is following the EU AI Act template rather than a US sectoral approach, which means firms already preparing for European risk-tier compliance will find the Chilean structure familiar. A company operating in both markets can reuse much of its EU classification and documentation work, while watching the Senate for changes that could shift specific thresholds or duties. The prudent posture is to prepare for a risk-tier regime in Chile without assuming the current text is final.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.