Skip to content
Data Protection

Data Privacy in the AI Era: An Ethiopian Perspective

By Haymanot & Advocates 7 November 2025 11 min read

Telecom

This series continues from the first instalment on data protection in Ethiopia, which explored the rights of data subjects under the Personal Data Protection Proclamation No.1321/2024 (PDPP). In this second part of the series, we focus on the intersection of Artificial Intelligence (AI) and data privacy, analyzing the trends, legal framework, emerging risks, and regulatory gaps that demand attention in this rapidly evolving field.

Legal Frameworks

Worldwide, the rapid advancement of AI has brought about significant transformation across industries, from finance and healthcare to legal, governance and digital services and Ethiopia is not an exception.  However, the widespread adoption of AI-driven technologies raises critical concerns about data privacy, especially in jurisdictions like Ethiopia, where regulatory frameworks are still evolving.

Efforts to regulate and oversee AI development have led to the adoption of several key policies and frameworks at global, continental, and national levels. Among these, UNESCO’s first global standard-setting instrument on AI ethics, endorsed by 193 countries, including Ethiopia is one milestone. Similarly, the OECD AI Principles, adopted by over 60 countries, and the EU AI Act, dubbed as the world’s most comprehensive AI regulation, is expected play a critical role in shaping AI governance.

At the continental level, the African Union (AU) adopted the “Continental Artificial Intelligence Strategy” in July 2024. This strategy is said to take an Africa-centric, development-oriented approach, aiming to harness AI for socioeconomic transformation. It underscores the importance of ethical, responsible, and inclusive AI practices, aligning with Agenda 2063 and the Sustainable Development Goals. At the national level, Ethiopia has been actively developing an AI strategy to drive socioeconomic progress. Following three years of extensive stakeholder consultations, the Ethiopian Council of Ministers unanimously approved the country’s first National AI Policy in June 2024. The Policy outlines Ethiopia’s ambition to integrate AI across multiple sectors, positioning the country as a leading AI hub in Africa by 2035. At the sector level, there are ongoing regulatory developments in AI regulation. The medical sector is an example, as the Ethiopian Food and Drug Authority issued guidelines in 2022 for software as a medical device (SaMD) and the use of AI-based medical devices (AI-MD).

While AI presents immense opportunities, concerns have been echoed ranging from job displacement to algorithmic bias. However, privacy and data protection risks remain, more or less, at the forefront of the discourse. Stakeholders continue to weigh the benefits of AI against its potential threats to data privacy, reflecting a broader conversation on the balance between technological advancement and fundamental rights. It is therefore a basic matter of policy and law to decide how we regulate the development of AI and address the potential risks it poses to privacy and data protection.

As covered in part one of this series, Ethiopia, too, has introduced comprehensive data protection rules with its PDPP, which offers comprehensive protections for data subjects, aligning with international standards. However, Generative AI represents a newer development, and both Ethiopian and international laws are not still sufficient in addressing the data privacy challenges it poses.

AI Policy

Ethiopia’s National Artificial Intelligence Policy, endorsed in June 2024, is a significant step in positioning the country as an AI hub in Africa by 2035. The Policy aims to leverage AI for socioeconomic growth, improving sectors like healthcare, education, agriculture, and national security. It prioritizes AI development over regulation, adopting a laissez-faire approach, though it highlights data protection and privacy concerns. Key areas include:

  • Data Ethics and Governance: Ethical guidelines for data usage are aligned with the PDPP’s data protection standards.
  • Healthcare and EMR: The policy emphasizes safeguarding electronic medical records and patient privacy, with additional regulatory measures outlined in the SaMD AI-MD Guideline.
  • Universal Identity Services: The policy supports AI integration in developing a national biometric identity system, subject to data processing rules under the PDPP and the Digital ID Proclamation.

Intersection of AI Regulation Privacy and Data Protection

AI Regulation

In general terms, AI involves machines replicating human cognitive abilities, including reasoning, decision-making, and task execution. The European Commission offers a more detailed definition, emphasizing the following aspects:

AI-based systems can be purely software-based, acting in the virtual world (e.g. voice assistants, image analysis software, search engines, speech and face recognition systems) or AI can be embedded in hardware devices (e.g. advanced robots, autonomous cars, drones or Internet of Things applications). AI systems require data—massive amounts of it—for training purposes.

Data is more frequently collected by machines from various sources. These include surveillance cameras, medical testing devices that capture information from the physical world, as well as digital systems that track online behavior and record transactions. Additionally, AI gathers data from biometric scanners, GPS and geolocation services, social media platforms, smart home devices, IoT-enabled appliances, voice recognition systems, autonomous vehicles, and satellite imaging. These diverse data sources enable AI to analyze vast amounts of information for various applications.

When it comes to AI regulation, two opposing perspectives on AI development emerge. The “techno-optimist” view champions the potential of technology and data to revolutionize industries and drive economic value, often downplaying privacy concerns and emphasizing the advantages of data analytics for businesses. Proponents of this approach argue that major AI companies should be free to develop AI rapidly and aggressively. While the transformative benefits of AI are undeniable, its reliance on continuous data collection—through websites, mobile devices, smart wearables, virtual assistants, drones, and facial recognition—raises significant concerns about privacy and data protection. Ignoring these risks poses a substantial threat to individuals’ rights.

From a policy and legal standpoint, the key characteristic of very large datasets, which defines them as big data from a functional perspective, is their potential for advanced analytics. This includes uncovering correlations, re-identifying individuals, and making predictions, often through AI driven techniques. The ability to process and analyze such vast amounts of data has significant implications for data protection as it raises concerns about privacy, consent, and the ethical use of personal information. The intersection of big data with AI driven analytics makes it particularly relevant for regulatory frameworks governing data security and individual rights.

Striking a balance between safeguarding personal data and maximizing its value requires a thorough examination of current and proposed legal frameworks, along with their real-world implications. When handled responsibly, personal data can drive economic growth, foster intellectual advancements, and contribute to the public good.

AI, Data Processing and the PDPP

The PDPP does not explicitly mention the term ‘AI.’ However, its scope covers data processing through automated means, including automated decision-making, which involves data processing by non-human intelligent systems. Moreover, several provisions within the PDPP are directly relevant to and applicable in the context of AI. It defines data as any information handled by automated systems. AI-driven processing falls within its scope.

Before discussing the regulation, it is important to note that AI-based data processing differs from traditional data processing methods and presents both emerging and foreseeable risks. Processing data through AI means automating the collection, analysis, and utilization of personal data using machine learning algorithms, predictive models, or decision-making systems. First, the scale and speed of data processing in AI systems is unparalleled. AI systems can analyze data at an unprecedented speed, often in real time, making it difficult to track, control, and regulate data flows. This results in lack of user control and profiling and inference risks. Second, the lack of explainability, commonly referred as the ‘black box effect’ is the attribute of AI powered data processing where the decision making of the system is complex and not explainable. This is a new advent from the traditional data processing that follows rule based algorithms where input-output relationships are clear. This has transparency risks and undetectable biases among others. Third, as AI powered systems require vast datasets for training and optimization, they appear to be antithetical to the ‘data minimization’ principle of data protection. This will result in excessive data collection, retention and data hoarding. Fourth, as opposed to traditional data processing models which can effectively anonymize data effectively, AI data processing systems, particularly deep learning models, can re-identify individuals even from anonymized datasets. Fifth, AI systems can analyze personal data, with the ‘black box effect’ and come up with automated decisions, as opposed to traditional systems which required human intervention and whose decision making process was transparent. This will pose the problem of unfair profiling, bias and perpetuating old bias. Sixth, as opposed to traditional data processing where companies and entities had clear control over who owned data and where it was stored, AI systems involve complex data ecosystems, including, third party data sharing and continuous learning models. These, are but some of the main novelties introduced by AI powered data processing.

The definition of Processing under the PDPP includes “an operation or set of operations performed on personal data, whether or not by automated means,” which can be related with how AI systems continuously collect, structure, store, and analyze vast amounts of personal data without direct human intervention. AI-driven systems go beyond simple data storage or retrieval, engaging in complex data combination, adaptation, and predictive analysis that can impact individuals’ privacy and rights.

Similarly, Profiling, as defined under the PDPP, refers to “any form of automated processing of personal data to evaluate certain personal aspects relating to an individual, including performance at work, economic situation, health, personal preferences, interests, reliability, behavior, location, or movements.” AI-based profiling systems, such as those used by targeted advertising algorithms in daily social media apps, fit into this definition. These AI models continuously analyze personal data to make predictions and decisions that can affect access to financial services, job opportunities, or even public services, often without transparency or user consent.

The PDPP defines Personal Data as any information that identifies a natural person, directly or indirectly. AI often extracts insights from personal data, linking behaviors, preferences, and digital footprints to individuals—sometimes even when data was not originally structured for identification. AI can process structured (databases, spreadsheets) and unstructured (text, images, voice) data, often in ways traditional processing cannot. Even if data is not in a structured filing system, AI models can infer patterns that lead to identification, making them subject to data protection rules.

AI data processing has significant implications for privacy and personal data protection, as highlighted in the six points above. For illustration purpose, let us see how AI models can re-identify individuals from raw data sets, which remains a recurring challenge to data protection and privacy.

Let us say a certain public hospital in Addis Ababa releases anonymous research data on diabetes medication usage trends for public health research. The data set states that 60% of diabetes patients in Addis Ababa use medication X. The average age of diabetes patients taking medication y is 55. The usage has increased by 20% in Addis Ababa over the past year. This data is fully anonymized and aggregated so under the definition of the PDPP it is not considered personal data because it cannot directly or indirectly identify individuals. Now, an AI system cross-references this dataset with public social media posts where individuals share their diabetes treatments, online patient forums discussing specific medications, pharmacy supply chain data showing purchases in specific areas. By combining these sources, the AI identifies a few specific individuals who are taking a certain diabetes medication. Now, the originally anonymous data becomes personal data. The hospital did not violate the law when releasing the anonymized data. But because AI can de-anonymize,it becomes personal data.

Data Protection & AI Data Processing in Ethiopia

The PDPP establishes six legal bases for processing personal data. These are consent, contract, legal obligation, vital interests, public interest/emergency, and legitimate interests. These same legal bases can be applicable for processing with AI systems. For instance the EFDA Guideline for SaM and MD-AI requires compliance with rules under national legislation.  However, the PDPP doesn’t mention AI, and doesn’t explicitly prohibit high-risk AI applications to be further regulated. For instance, the EU AI Act prohibits specific AI tools that could potentially result in discrimination in data analysis.

Lawful, Fair, and Transparent Processing

The first principle under the PDPP requires that personal data be processed lawfully, fairly, and in a transparent manner. AI-powered systems such as AI-driven ad targeting social media apps often operate with limited transparency. Ethiopian users may not fully understand how their data is collected, used, or shared with third parties. The PDPP has made it possible that if the personal data has been processed unlawfully, individuals have the right to claim for such data to be deleted as soon as possible without any payment. However, there is no requirement specific for baseline level of transparency for AI models.

A notable reference regarding automated decision making process is the right to obtain human intervention, the right not to be subject to a decision based solely on automated processing and so on.

Purpose Limitation

The second principle states that personal data must be obtained for explicit, specified, and lawful purposes and not further processed in a way that is incompatible with those purposes. However, repurposing is one of the challenges automated data processing has. Social media apps and everyday apps will collect data and process it towards another purpose.

Data Minimization

‘Privacy by Design’, a proactive approach to data protection that integrates privacy into the design of systems and processes is advanced as the best way to implement data minimization. Data Minimization, the third principle requires that personal data be adequate, relevant, and not excessive. However, AI systems collect large amounts of user data, often beyond what is necessary. Many AI-driven businesses, particularly those in advertising, social media, and analytics, have incentives to maximize data collection rather than minimize it. AI models, especially those based on machine learning, improve their accuracy and effectiveness when trained on large datasets.

Storage Limitation

This principle under PDPP requires that personal data be kept only as long as necessary. There is also a data deletion mandate under the. A data subject has the right to erasure of personal data where the data is no longer necessary in relation to the purpose for which they were collected or processed.

Regulatory Gaps

Ethiopia currently does not have AI-specific legislation. While the PDPP emphasizes the importance of securing personal data throughout its lifecycle, AI systems present unique risks that require additional security measures beyond traditional data protection practices.

Moreover, while the PDPP mandates the implementation of technical and organizational measures, it fails to identify and establish high-risk security protocols for high-risk AI systems. In contrast, the EFDA Guideline provides more comprehensive requirements for AI-MD, including addressing the risks of continuous learning in AI systems and mandating effective control over the learning process and evolving performance of AI-MD systems.

Recommendation

Given these gaps, it is essential to introduce measures that specifically address the unique characteristics of AI models and the challenges they pose to privacy and data protection. These measures should take a two-pronged approach. First, there is a need to enact a comprehensive framework legislation that provides an overarching regulatory structure for AI systems, ensuring that privacy and data protection are effectively safeguarded across all sectors. Second, sector-specific directives or regulatory guidelines, such as those exemplified by the EFDA, should be introduced to address the particular risks and requirements associated with AI in different industries. This dual approach would ensure both broad and targeted protections for AI systems, promoting a more secure and privacy-conscious environment.

This article is general information and not legal advice. Contact us about your specific situation.

H&A

Published by

Haymanot & Advocates

Legal update · Data Protection practice