Responsible Artificial Intelligence Governance
The adoption of Artificial Intelligence (AI) solutions has transformed the way organizations analyze data, make decisions, and provide services. This potential, however, comes with certain requirements: the closer these technologies come to making decisions that affect people, the more important it becomes to ensure that they operate ethically, transparently, and responsibly, without reproducing biases, exposing personal data, or causing unintended consequences for society and the environment.
Aware of this trend, Copel treats responsible Artificial Intelligence as a governance issue, not merely a technological one. The Company understands that trust and sustainability go hand in hand, which is why it structures the development and use of AI based on formal guidelines, well-defined responsibilities, and controls that accompany each stage of the solutions’ lifecycle.
Approval of the vision and strategy for the use of Artificial Intelligence at Copel rests with the Board of Directors (CAD). Based on this direction, the Superintendence of Technology and Digital Transformation, which reports to the Vice Presidency of Strategy, New Business, and Digital Transformation and operates with the approval of the Joint Executive Board (Redir), leads executive management and supports the implementation of the strategy. Oversight of internal controls and risk management related to this area falls under the purview of the Statutory Audit Committee (CAE), which reports directly to the CAD.
At the operational level, Copel maintains strategies, policies, processes, and tools that are periodically reviewed to guide the design and implementation of its AI solutions. The goal is to safeguard human oversight, equity, transparency, and the reliability of the systems, as well as to ensure a clear attribution of responsibility for the decisions supported by them, conditions that underpin stakeholder trust.
Information Technology and Cybersecurity Policy (NPC 0301)
This commitment is embodied in a set of practices and safeguards, guided by the Information Technology and Cybersecurity Policy (NPC 0301), approved by the CAD, which establishes guidelines for the responsible use of Artificial Intelligence within the Company. Structured human oversight is maintained for critical operational decisions, and the portfolio of solutions remains centralized under the coordination of the Vice Presidency of Strategy, New Business, and Digital Transformation, with approval of compliance regarding privacy, risks, and legal aspects. The Company prohibits the use of solutions based on behavioral manipulation, social scoring, or remote biometric identification outside of legally authorized circumstances, and restricts access to sensitive capabilities to authorized personnel. In addition to these measures, the Company conducts recurring evaluations of models to identify and mitigate biases, continuously monitors their performance over time, and trains employees in the ethical, safe, and responsible use of Artificial Intelligence.
Furthermore, the processing of personal data in the context of AI solutions complies with Federal Law No. 13,709/2018, the General Personal Data Protection Law (LGPD), and the guidelines of the Copel Privacy and Personal Data Protection Policy (NPC 0322), ensuring respect for the rights of data subjects.
Copel maintains an appeals process for users or third parties affected by decisions or outcomes resulting from Artificial Intelligence, allowing them to challenge, request a review of, and seek clarification on AI-supported decisions. This process is carried out through the Company’s Reporting Channel, which includes a specific category for reports related to Artificial Intelligence.
Guidelines for the Responsible Use of Artificial Intelligence
The Information Technology and Cybersecurity Policy (NPC 0301) establishes the guidelines that govern the development, acquisition, and use of artificial intelligence solutions at Copel, as detailed below.
- Security of Artificial Intelligence systems. Promote cybersecurity measures throughout the development, acquisition, and use of AI systems, including security testing, secure coding practices, and control mechanisms—such as guardrails—designed to prevent attacks, manipulation, and the generation of harmful or inappropriate content.
- Data protection and lifecycle controls. Establish cybersecurity and personal data protection controls as an integral part of the development, procurement, and lifecycle of AI applications and systems, ensuring strict compliance with legislation and that the information processed is protected in accordance with its classification and risk exposure.
- Equity and bias mitigation. Establish processes that ensure ethical, fair, and non-discriminatory behavior in the development, procurement, and use of AI, identifying, mitigating, and auditing potential biases and prioritizing, where applicable, diverse and representative training datasets.
- Human-in-the-Loop supervision. Ensure documented human supervision in AI systems that support or execute critical or high-risk decisions, guaranteeing the ability for humans to intervene, review, and override decisions in order to prevent irreversible autonomous actions.
- Transparency and explainability. Promote the transparency and explainability of AI systems by clearly disclosing their use, capabilities, and limitations; providing, when technically feasible, understandable justifications for the results generated; and explicitly informing users when they are interacting with an AI system.
- Limits and authorized scope. Clearly define and document the limits, authorized scope, and capabilities of AI systems, implementing technological safeguards and permitted-use policies that reduce the risk of operation outside the intended purpose or of unintended expansion.
- Sustainability. Prioritize sustainability in the development, acquisition, and operation of AI models and infrastructure, promoting computational optimization techniques, energy efficiency, and the preferential use of renewable energy sources.
- Prohibitions on Unacceptable Use. Strictly prohibit the development, deployment, or use of AI systems that involve manipulative or deceptive techniques, the exploitation of vulnerabilities, social scoring practices, or unauthorized biometric identification systems.
Life Cycle of Artificial Intelligence Models
Copel follows a structured life cycle for its Artificial Intelligence models, designed to ensure reliable, ethical, secure, and sustainable solutions over time. This cycle is organized into sequential and iterative phases, from problem definition and data curation to development, evaluation, deployment, and monitoring, each with defined responsibilities and its own control mechanisms. Non-negotiable governance requirements apply to all stages, such as traceability, auditable documentation, independent review, regulatory compliance, and the treatment of ethics as a non-negotiable value.
Fairness assessment and bias mitigation. Prior to deployment, models undergo evaluation, validation, and bias auditing, using equity metrics and a documented methodology, to identify performance disparities among subgroups and apply the necessary corrections.
Drift detection and performance maintenance. AI models can lose effectiveness over time as data patterns or external conditions change, a phenomenon known as model drift. To address this, Copel continuously monitors the performance of models in production, using periodic statistical tests and alerts based on degradation thresholds. Once a performance decline is confirmed, retraining triggers are activated, followed by a new bias audit before the model is carefully returned to production, ensuring the solution remains reliable and relevant.
Sustainability and Computational Efficiency in AI
Copel considers sustainability a key criterion in the acquisition and use of infrastructure and artificial intelligence models, prioritizing solutions that incorporate energy efficiency and computational optimization. To this end, the Company uses Google’s data center infrastructure and Artificial Intelligence services; Google is a provider that adopts energy-efficiency and resource-optimization procedures in its operations, contributing to the reduction of emissions associated with digital processing.
Copel implements initiatives that utilize artificial intelligence to generate measurable environmental and social benefits, in line with its sustainability goals. Among these, the vegetation management tool for transmission line corridors, developed in partnership with the Dutch company Overstory, stands out. It uses 3D satellite imagery and artificial intelligence to identify potential risks of interference with the power grid. The impacts are quantified using indicators related to the height, type, and condition of the trees, enabling the identification of high-risk areas and their subsequent validation during field inspections. This monitoring makes it possible to assess the solution’s contribution to reducing outages, increasing the resilience of the power infrastructure, and optimizing the use of operational resources. Thus, the initiative demonstrates measurable benefits resulting from the responsible application of artificial intelligence within the company.
Another initiative is HarpIA Guardian, a solution developed through an ANEEL research and development (R&D) project carried out by SGRIDD Tecnologia and USP, in partnership with Copel. The tool uses artificial intelligence, intelligent image processing, and edge computing to support workplace safety during field operations by identifying risky situations and providing real-time decision-making support. By enhancing teams’ situational awareness and helping to prevent accidents, the solution generates significant social benefits, promoting safer work environments and protecting workers’ well-being.
In the environmental sphere, the Target Selector stands out, an AI solution designed to detect irregular procedures that directs field teams more effectively, reducing travel. The benefits of this initiative can be measured through the indicator of associated CO₂ emissions reduction, which demonstrates the solution’s positive impact. The Company seeks to continuously improve the measurement of these effects in order to demonstrate, through metrics and indicators, the contribution of its AI tools to achieving environmental and social outcomes.
Training on the Ethical and Safe Use of Artificial Intelligence
Copel promotes ongoing training for its employees on the ethical use, data privacy, and security of Artificial Intelligence systems, recognizing that the responsible adoption of these technologies depends as much on technical controls as on the preparedness of the people who use them. This training raises awareness of risks, biases, and best practices, fostering an organizational culture of responsibility in the adoption of AI and enhancing employees’ ability to recognize and appropriately address the ethical and security issues involved.
The training sessions are conducted through the KnowBe4 platform, which the Company uses to develop skills in security and the responsible use of technology. Training takes place on a regular basis, with sessions already completed and new cycles planned, ensuring that employees remain up to date as artificial intelligence technologies evolve and best practices associated with their use develop.
In addition, the Company conducts periodic training in partnership with Gentrop, Inteli, and other Google partners, expanding the scope and depth of its teams’ training. Employees also have access to learning platforms such as Google Skill Boost and Udemy, which offer up-to-date training tracks and support the continuous development of skills related to Artificial Intelligence and related technologies.
