Artificial intelligence is not changing the purpose of impact assessments, but it is changing the risks organisations need to assess. As AI becomes embedded in business processes, organisations are increasingly required to consider not only privacy risks but also broader impacts on fundamental rights, governance and decision-making. Understanding how Data Protection Impact Assessments (DPIAs) and Fundamental Rights Impact Assessments (FRIAs) fit into this evolving landscape is becoming an essential part of responsible AI governance.
Artificial intelligence is changing the questions organisations need to ask
For many organisations, DPIAs have become a familiar part of risk management and compliance. The GDPR framework has been in place for years, and data protection professionals are generally comfortable assessing risks related to personal data processing.
Even though artificial intelligence introduces new technologies, risks and challenges, it does not fundamentally change why we conduct impact assessments in the first place. Organisations are still expected to identify risks before they materialise and implement appropriate safeguards where necessary. What AI has changed is the nature of those risks.
As organisations increasingly deploy AI-enabled tools, use generative AI applications and operate within highly connected digital environments, traditional impact assessment methods are being applied to situations that look very different from those envisioned when many organisations established their DPIA processes.
From data processing to information generation
Historically, data protection assessments have focused on questions such as:
- What personal data is being processed?
- Who has access to it?
- How long is it retained?
- Can the processing be justified and secured?
These questions remain important. However, AI introduces an additional dimension.
Many AI systems do not simply store, transfer or display information. They generate outputs, create summaries, infer patterns and support or influence human decision-making. As a result, organisations increasingly need to assess not only the data entering a system but also the information emerging from it.
The key risks are often no longer limited to data collection or disclosure. They may arise from inaccurate outputs, misleading recommendations, excessive reliance on automation or the unexpected combination of information from multiple sources.
Looking beyond privacy
Privacy and personal data protection remain central considerations in any impact assessment involving personal data. At the same time, AI has contributed to a broader discussion about the potential impact of technology on individuals and society.
Questions relating to fairness, discrimination, transparency and effective human oversight have become mainstream governance concerns. In many organisations, these issues are no longer discussed only by privacy teams. Legal, compliance, information security, HR and business stakeholders are increasingly involved as well.
This broader perspective is reflected in the growing importance of fundamental rights considerations and, under certain circumstances, the requirement to conduct a Fundamental Rights Impact Assessment (FRIA) under the EU AI Act.
It is important to note that a FRIA is not a universal requirement for all organisations using AI. Under Article 27 of the AI Act, the obligation applies only to certain deployers of high-risk AI systems, including public authorities, bodies governed by public law, certain private entities providing public services, and organisations deploying specific AI systems used for creditworthiness assessments or life and health insurance decisions. Nevertheless, organisations outside the scope of the legal obligation may still find the FRIA methodology useful when assessing broader impacts on individuals and society.
Risk assessment is becoming more multidisciplinary
One interesting development is that impact assessments are gradually moving beyond their traditional silos. A DPIA, information security assessment, AI governance review, supplier assessment and a FRIA may all examine different aspects of the same project.
Although the legal requirements and assessment criteria differ, many of the underlying questions are remarkably similar:
- What is the intended use case?
- Who may be affected?
- What could go wrong?
- What safeguards are available?
- How will risks be monitored and managed over time?
This is encouraging organisations to adopt a more integrated approach to risk assessment and governance rather than treating each assessment as a separate compliance exercise.
The most important question remains unchanged
Despite the rapid development of AI technologies, one principle remains surprisingly constant. Effective impact assessments start with people, not technology.
Organisations often begin by asking whether a particular AI tool requires a DPIA, a FRIA or some other form of assessment. While these are important questions, they are rarely the best starting point.
A more useful approach is to understand the use case itself: who may be affected, how decisions are being supported, what information is involved and what the potential consequences could be if something goes wrong.
Whether the assessment ultimately takes the form of a DPIA, a FRIA or another risk assessment is often a secondary question.
A changing discipline
The rise of AI is not replacing traditional impact assessments. Instead, it is expanding them. Data protection risks remain important, but organisations increasingly need to consider how AI influences information flows, decision-making, fundamental rights and governance processes more broadly.
For data protection and compliance professionals, this means that impact assessments are becoming less about documenting processing activities and more about understanding the real-world consequences that technology may have for individuals. In that sense, the purpose of impact assessments remains exactly the same as it has always been: helping organisations identify and manage risk before harm occurs.
