Bill No. 18,820-19, introduced in the Senate on October 28th, proposes amendments to Law No. 21,663, Chile’s Cybersecurity Framework Law, establishing new obligations for private institutions subject to the Law that use artificial intelligence (AI) Agents capable of taking actions within their networks and IT systems.  

What does the bill seek to achieve? 

The bill seeks to bring AI Agents directly within Chile’s cybersecurity regulatory framework by establishing measures to prevent them from operating beyond the permissions granted by the organizations that use them. Unauthorized actions by these Agents could compromise sensitive information, disrupt processes or have an impact on other institutions.  

The Cybersecurity Framework Law establishes obligations relating to the prevention, reporting and response to cybersecurity incidents, together with specific requirements for Operators of Vital Importance (Operadores de Importancia Vital, or OIVs, in Spanish). Building on this framework, the bill introduces specific requirements for AI Agents, given their ability to act without human intervention in each individual action and to delegate tasks or activate other Agents.  

Among other changes, the proposal would incorporate a definition of an AI Agent into the Cybersecurity Framework Law. An AI Agent would be understood as a system capable of selecting and executing actions on networks and IT systems to achieve specific objectives without requiring human intervention for each action. The definition would also cover systems in which certain operations require prior human authorization.

In practical terms, this includes systems capable of accessing information, modifying files, using credentials or assigning tasks to other AI Agents.

The proposal would also introduce the concepts of a Loss of Control Incident, which occurs when a cybersecurity incident results from an AI Agent taking actions that exceed its permissions or authorized scope, and a Loss of Control Near Miss, which occurs when an AI Agent bypasses a measure intended to restrict its permissions or scope of operation, but is contained by another measure before causing harm or compromising any of the elements of a cybersecurity incident.

What new obligations would companies have? 

Private institutions subject to the Cybersecurity Framework Law under Article 4 would be required to implement a series of controls when using AI Agents. The main requirements include:

• Assigning each AI Agent a verifiable digital identity so that its actions can be individually identified. 

• Defining and recording its permissions in advance, including who authorized its operation, which resources it may access, which actions it may perform and for how long. 

• Recording the delegation of tasks to other AI Agents and ensuring that delegated agents do not obtain permissions exceeding those originally authorized. 

• Keeping logs of the Agent’s activities so that it is possible to reconstruct what the Agent did, when it acted and under which authorization. 

• Implementing monitoring and control mechanisms capable of detecting deviations and suspending or revoking access. 

• Implementing mechanisms to safely stop an AI Agent when it operates beyond its authorized permissions. 

• Keeping internal records of Loss of Control Near Misses.

Private institutions classified as OIVs would also be required to:

• Maintain an up-to-date register of AI Agents with access to production systems. 

• Assess the controls applicable to AI Agents before deploying them. 

• Conduct new assessments following material changes. 

• Perform periodic security testing based on the appropriate level of risk, including assessments of the effectiveness of existing controls and the possibility that those controls may be bypassed or fail to operate as intended.  

What should companies review? 

As the bill is still at an early stage of the legislative process, its provisions may change. However, under the proposal as currently drafted, companies that already use AI Agents or are considering doing so should review which Agents they use, the systems and credentials those Agents can access, and whether they have appropriate records and mechanisms in place to monitor and stop them when necessary.

These controls should be proportionate to the Agent’s level of autonomy, the permissions granted to it, the sensitivity of the information it can access and the criticality of the systems involved.

When AI Agents are provided by third parties, companies should also review the relevant contracts. Under the current proposal, outsourcing the service would not release a company from its obligations under the Cybersecurity Framework Law.

When would the new obligations take effect? 

The bill provides for a 12-month implementation period following the publication of the eventual law, giving organizations time to adapt to the new requirements.

For more information, please contact Fischer y Cía.’s Corporate team. 

Catalina Wastavino 
Senior Associate, Fischer y Cía.