Personal Data Protection Authority Publishes the Document Titled “Agentic AI”

The Personal Data Protection Authority (Authority) published a document titled “Agentic AI” considering recent developments in artificial intelligence technologies. The document discusses the main characteristics of agentic AI systems, the AI agents used within these systems, their potential areas of application, and the risks they may pose in terms of the protection of personal data. The study aims to provide a general assessment of data protection considerations that may arise throughout the lifecycle of agentic AI systems. 

The key points are summarized below:

• Agentic AI systems are defined as integrated structures that can assess environmental conditions, perform planning, and initiate actions with a certain degree of autonomy by adapting to changing circumstances in order to achieve specific goals.
• These systems generally operate through “AI Agents”, which are described as software components capable of perceiving their environment, making decisions in line with specified objectives, and performing the necessary actions to achieve those objectives.
• It is indicated that agentic AI systems may have potential use cases in areas such as research and development processes, customer support services, financial analysis, decision-support systems in the healthcare sector, and incident response.
• The goal-oriented and multi-step operational structure of such systems may create risks in terms of predictability, transparency, and accountability, and erroneous or biased outputs may also lead to cascading effects.
• The collection of personal data from multiple data sources and its processing through multiple stages in the use of agentic AI systems may create additional risks with respect to data minimization, purpose limitation, and data security.
• In this context, it is recommended to ensure human oversight, develop accountability mechanisms, safeguard data accuracy, clearly define roles and responsibilities, and adopt a privacy by design approach.