From Conversational Agents to Tax Assistants: A roadmap towards simpler, traceable, and accessible tax compliance.

  1. Introduction

In a recent post on this blog[1], I introduced the fascinating concept of Artificial Intelligence (AI) Agents, inviting readers to reflect on how they could mark a new frontier in tax management.

This post proposes to explore how we can apply the sophistication of AI Agents in a more concrete way regarding the aspect of action related to the assistance and facilitation of tax compliance[2], through Tax Assistants.

These assistants represent a specialized and powerful application of conversational Agents, which could be specifically designed to navigate and simplify the complex tax ecosystem, which faces significant challenges (very technical, complex and voluminous regulations, proliferation of obligations and additional duties to the payment of taxes to be fulfilled through various channels and mechanisms, digital literacy gaps and high costs of personalized attention, among others). This is where the maturation of conversational AI, precisely from the perspective of AI agents, presents us with a unique opportunity to transform compliance, with challenges and challenges to consider.

  1. From “Search” to “Solve”: The Agentic[3] shift in the digital interaction.

Taxpayers’ digital interaction with Tax Administrations is limited to the use of conversational agents or chatbots[4] that answer simple questions. Now, Tax Assistants, as an evolution of conversational agents, transform this approach: they understand the user’s intention, maintain the context of the conversation, integrate multiple sources, and execute specific actions such as completing forms or initiating procedures.

This change replaces fragmented interfaces with integral conversational experiences, where the system not only guides, but validates data, records actions and concludes processes under institutional control.

Recent advances in conversational AI, such as agentic functions[5] of ChatGPT or the autonomous approach of “Perplexity Comet”[6], confirm this trend towards systems capable of solving tax needs in an active and automated way.

  1. Potential Applications for the Tax Assistant

The Tax assistants can provide immediate value in various areas of tax compliance:

  • Comprehensive management of the tax cycle: The assistants guide the registration for taxes, data updating and filing of returns, automating key steps such as the upload of information and documentary validation.
  • Explanatory normative, doctrinal, and jurisprudential orientation[7]: They offer answers based on articles, paragraphs, and specific resolutions, showing the citations and current versions. The traceability of the reasoning (which rule consulted, which section applied, how was it interpreted) becomes fundamental to preserve legal certainty.
  • Multi-channel accessibility: They operate by text or voice, on multiple devices, and keep the interaction history, which expands access to people with technological or cognitive barriers. You can always escalate to a human agent in more complex situations.
  1. Vertical Tax Assistants: Close domain, greater control.

A particularly effective approach is the design of vertical Assistants. These systems’ training, being a type of conversational agent, circumscribe them to a specific legal-regulatory domain, with controlled sources and dated versions.

They use augmented retrieval techniques to minimize unsubstantiated responses, prioritizing auditability (each response relates to its normative foundations) and the reduction of hallucinations (the retrieval of official documents is privileged, and the text generation conditioned to those references).

In addition, they ensure complete traceability (each interaction leaves a verifiable trace of consulted data, applied rules and suggested decisions) and promote interoperability through the use of APIs and standards that allow internal and external sources to connect securely. Focusing on a narrow domain facilitates governance, limits systemic risk, and improves the quality of the responses of these Assistants.

  1. Protection of Personal Data: Principles and Safeguards for the Tax Assistant

The deployment of Tax Assistants requires strict compliance with data protection regulations (including the duty of tax secrecy) and robust ethical guidelines. Their minimum pillars include:

  • Legality and specific purpose: The data processing must have a clear legal basis and a specific purpose, avoiding the use of tax data for unauthorized purposes.
  • Minimization and proportionality: We collect and process only the data strictly necessary for the procedure, avoiding unnecessary accumulations that increase the risk surface.
  • Transparency and informed consent: Users should understand what data we use, for what purposes, for how long and with whom we share them. If consent is the legal basis, it must be express, unequivocal and revocable.
  • Security and confidentiality: Systems must implement encryption, access segmentation, event logging, periodic penetration testing and incident response plans.
  • Individual Rights: There should be clear channels for people to access, rectify, delete, oppose and, where appropriate, port their data, as well as challenge automated decisions that affect their rights.
  • Impact assessments: Before deployment, it is advisable to conduct impact assessments on data protection and taxpayer rights, reviewing them periodically as the system evolves.
  1. Governance, Risks and Meaningful human control in the context of Tax Assistants

The adoption of Tax Assistants does not imply a blind delegation of responsibilities. We must implement a framework of algorithmic governance that would include:

  • Clearly defined roles and responsibilities.
  • Continuous evaluation protocols for accuracy, robustness, biases, explainability and security.
  • Full traceability of automated decisions and actions.
  • Meaningful human control, with real capacity for review, correction, and veto.
  • Update policies (model, data, rules) and documented versioning.
  • Contingency plan in case of failures, errors, or security incidents.
  • Technological independence strategy to mitigate risks of extreme dependence on suppliers.
  1. Suggested Roadmap for the Implementation of Tax Assistants

We propose the following gradual path to implementation:

Phase 1: Institutional preparation

  • Inventory of data, processes, APIs, and internal capabilities.
  • Definition of applicable AI principles (data protection, explainability, security, human control) for the attendees.
  • Interdisciplinary training (technology, legal, tax, ethics).

Phase 2: Controlled pilots

  • Selection of low-risk and high-volume cases (frequently asked questions, simple registration, data of simplified schemes) for the testing of assistants.
  • Closed monitoring of accuracy, errors, response times, and referrals to human operators.
  • Legal and data protection impact assessments prior to escalation.

Phase 3: Gradual escalation

  • Extension of the functionalities of the assistants with deeper integrations into transactional systems.
  • Periodic audits, transparency reports, and accountability mechanisms.

Phase 4: Continuous improvement and supervised automation

  • Progressive automation of more complex procedures through the assistants, always with significant human control.
  • Constant updating of models, data, and governance policies.
  • Regular review cycles of the technical, legal, and social impact.
  1. Conclusions

Tax Assistants, as the natural evolution of conversational agents, represent a new frontier for today’s tax management. Their promise does not lie in replacing institutions, but in bringing the system closer to the citizens, simplifying compliance, and raising the quality and traceability of the service. To fulfill that promise, it is essential to align technological design and institutional protection: protect data, ensure explainability, sustain human control, continuously audit and advance in stages.

The immediate future of the link between taxpayers and tax systems will not only be to search and read, but to dialogue, understand and complete processes within safe, auditable, and responsible environments with the help of these Assistants. The key is to balance innovation and prudence, so that AI becomes a true facilitator of tax compliance.

 

References:

[1] Porporatto, p. (2025, June 14). Artificial Intelligence Agents: A new frontier of intelligent tax management? The CIAT Blog. Retrieved from https://www.ciat.org/agentes-de-inteligencia-artificial-una-nueva-frontera-de-la-gestion-tributaria-inteligente/

[2] The other side aims at preventing, detecting, and regularizing non-compliance as a deterrent mechanism to induce compliance (Díaz Yubero, F., & Porporatto, P. (2020, February 3). Tax administrations in the Twenty-First Century CIAT Blog. https://www.ciat.org/las-administraciones-tributarias-en-el-siglo-xxi/).

[3] The term “agentic” refers to artificial intelligences designed to function as independent agents. According to John Roese, global director of technology and AI at Dell Technologies.

[4] Almost 75% of the advanced tax administrations at the global level already have chatbots to serve the taxpayers (Porporatto, p. (2025, June 23). Tax compliance with disruptive technologies: What are advanced tax administrations using? Mercojuris. https://mercojuris.com/cumplimiento-tributario-con-tecnologias-disruptivas-que-estan-utilizando-las-administraciones-tributarias-avanzadas/)

[5] Introducing Agent ChatGPT: A bridge between research and action:  https://openai.com/es-419/index/introducing-chatgpt-agent/

[6] Introducing Comet: Browse at the speed of thought – Perplexity Blog: https://www.perplexity.ai/hub/blog/introducing-comet

[7] In 2024, the TFN presented its innovative AI-Assisted Jurisprudence Search System (https://jurisprudenciatfn.mecon.gob.ar/) incorporating advanced Natural Language Processing technologies and specific filters that allow semantic and specialized searches by various criteria. In addition, it is capable of automatically generating fault summaries (Porporatto, e.g., A. (2024, December 8). The Nation’s Tax Court in the digital age: 64 years of creative innovation. Mercojuris. https://mercojuris.com/el-tribunal-fiscal-de-la-nacion-en-la-era-digital-64-anos-de-innovacion-y-vanguardia-dr-pablo-a-porporatto-actual-juez-del-tribunal-fiscal-de-la-nacion/).

Leave a Reply

Your email address will not be published.

CIAT Subscriptions

Browse through the site without restrictions. Consult and download the contents.

Subscribe to our electronic newsletters:

  • Blog
  • Academic offer (Only in spanish)
  • Newsletter
  • Publications
  • News alert

Activate subscription

CIAT Members

Representatives, Correspondent and Authorized staff (TA)