Agentic AI: UK Data Protection Risks and Considerations for Businesses

The rapid evolution of artificial intelligence (AI) continues to reshape the business landscape, with agentic AI poised to be the next transformative leap. The Information Commissioner’s Office (ICO) recently highlighted the potential of agentic AI to revolutionise commerce, from digital personal assistants to autonomous shopping agents. The ICO’s update makes clear that the benefits of agentic AI must not come at the expense of data privacy. The regulator will be actively monitoring developments and expects organisations to be proactive in embedding data protection by design and by default.

As these systems become increasingly capable of independent decision-making and action, the data protection risks they present are both heightened and novel. While the familiar principles of UK data protection law remain foundational, agentic AI introduces new complexities that organisations must address.

1. What is ‘agentic AI’ and how is it used?

Agentic AI refers to artificial intelligence systems endowed with the capacity to act autonomously – making decisions, interacting with their environment, solving problems in real time and engaging in reasoning and planning. Unlike traditional AI, which typically operates within defined parameters and requires explicit prompts, agentic AI can anticipate needs, initiate actions and negotiate outcomes on behalf of users.

The applications of agentic AI are broad and rapidly expanding. Examples of current and future use cases include using agentic AI to:

• automate client onboarding, monitor transactions for compliance and even negotiate loan terms or investment options;
• autonomously execute trades, rebalance portfolios or source and negotiate financing arrangements;
• create digital personal assistants managing diaries and travel; or
• create AI agents handling procurement and contract negotiations.

As agentic AI becomes embedded in core business processes, the data protection implications become more acute and multifaceted.

2. Key data protection risks unique to agentic AI

While agentic AI builds on the same data-driven foundations as traditional AI, its autonomy and capacity for real-time decision-making introduce additional risks. Below, we have categorised some of the key risks and outline some of the principal UK data protection considerations for each.

a) Autonomous decision-making

Agentic AI systems are designed to operate independently, making their own decisions and taking actions without human prompts, which means they can initiate new tasks or processes dynamically and unpredictably.

(i) Scope creep and purpose limitation
Agentic AI’s ability to act independently increases the risk of “scope creep” – where the system processes personal data for purposes beyond those originally envisaged. Organisations must ensure that agentic AI systems are designed with robust purpose limitation controls and that any expansion of processing activities is subject to fresh data protection impact assessments (DPIAs).

(ii) Data minimisation in dynamic environments
Agentic AI may collect and process data in real time, adapting to new contexts and objectives. Ensuring compliance with the data minimisation principle requires ongoing monitoring and technical controls to restrict data collection to what is strictly necessary for each autonomous task.

(iii) Lawful basis for autonomous actions
Agentic AI may initiate processing activities that were not foreseen at the point of deployment. Organisations must ensure that all potential processing activities have a clearly identified lawful basis under the UK GDPR and that mechanisms are in place to review and update these as the system evolves.

Agentic AI may also infer or use special category data (e.g., health, ethnicity) in unexpected ways, even if not intended by the original purpose, raising the need for enhanced safeguards and lawful bases for processing.

(iv) Facilitating data subject rights
Agentic AI’s autonomous actions may complicate the exercise of data subject rights (e.g., access, rectification, erasure, objection). Organisations must ensure that rights can be exercised effectively, including the ability to intervene and override AI decisions where necessary.

(v) Human-in-the-loop controls
To maintain trust and compliance, it is critical to retain meaningful human oversight over agentic AI systems, particularly where decisions have legal or similarly significant effects.

(vi) Accuracy and hallucinations
Agentic AI systems may generate inaccurate yet plausible information – so-called “hallucinations” – which, due to their autonomous operation, can propagate through multiple decisions and interconnected systems. This creates a risk of embedded errors and potential harms, particularly where outputs are relied upon for consequential business or customer decisions. Organisations should implement robust validation, monitoring and escalation processes to detect and correct inaccuracies promptly and ensure that human oversight is maintained where outputs could have significant legal, financial or reputational impact.

(vii) Retention
Determining appropriate retention periods for personal data processed by agentic AI requires careful consideration of the nature and frequency of the tasks performed by the agent. Where agentic AI is engaged in repetitive or ongoing activities (such as continuous monitoring, regular procurement or recurring financial analysis) there may be a legitimate need to retain certain data for longer periods to ensure operational continuity and efficiency. However, organisations must balance these operational needs against the data minimisation and storage limitation principles under UK GDPR, ensuring that personal data is not kept for longer than necessary. This may involve implementing task-specific retention policies, automated deletion protocols and regular reviews to assess whether continued retention remains justified as the agent’s functions evolve.

b) Transparency and explainability

Agentic AI often uses complex reasoning and adapts its behaviour in real time, making it inherently more difficult to trace, understand or explain the logic behind its decisions compared to more static or rule-based AI.

(i) Fair processing and meeting the obligation to inform
The opacity of agentic AI’s decision-making processes can make it challenging to provide clear and intelligible information to data subjects in accordance with fair processing obligations under the UK GDPR. Organisations must develop transparency mechanisms, which may include real-time notifications and accessible explanations of how and why agentic AI systems make decisions. Where agentic AI generates inferences or creates new personal data (such as predictions, classifications or profiles) organisations must ensure that data subjects are appropriately informed about the existence and nature of this inferred data, the purposes for which it will be used and their rights in relation to it.

(ii) Algorithmic explainability
Given agentic AI’s capacity for complex, unsupervised reasoning, ensuring meaningful explainability is critical. This may require investment in explainable AI (XAI) techniques and the ability to audit and reconstruct decision pathways, particularly where decisions have significant effects on individuals.

c) Security and risk of unauthorised actions

Because agentic AI can access systems and execute actions on its own initiative, it increases the risk that unintended or unauthorised activities could occur, either through system error or external manipulation, without immediate human oversight.

(i) Real-time access to sensitive data
Agentic AI may require access to sensitive personal data (e.g., financial information, health records) to perform its functions. Organisations must implement advanced access controls, continuous monitoring and incident response protocols to mitigate the risk of unauthorised access or misuse.

(ii) Safeguarding against malicious manipulation
The autonomy of agentic AI increases the attack surface for adversaries seeking to manipulate or subvert the system. Regular security testing, adversarial robustness assessments and prompt patching of vulnerabilities are essential.

3. Practical steps for deploying agentic AI responsibly

a) DPIAs and internal policy documents

Organisations considering the deployment of agentic AI should adopt a structured approach to risk management. This includes conducting thorough DPIAs at the earliest stages, establishing clear governance frameworks and ensuring ongoing staff training on the unique challenges posed by agentic AI. Regular reviews and updates to policies and procedures will be essential as the technology and regulatory expectations evolve.

b) Clarifying controller and processor roles in the agentic AI supply chain

The autonomous and interactive nature of agentic AI makes it increasingly challenging to determine who is acting as a data controller or processor at any given stage. As agentic AI systems independently interact with multiple platforms, vendors and data sources, organisations must carefully map data flows and contractual relationships to ensure that roles and responsibilities are clearly defined and documented. This may require revisiting and updating agreements, as well as maintaining ongoing dialogue with technology partners, to ensure compliance with UK GDPR accountability requirements and to avoid gaps in governance as agentic AI systems evolve.

c)Third party contracts

Agentic AI systems often rely on cloud-based infrastructure and may process data across multiple jurisdictions. Organisations must ensure that any international transfers of personal data comply with UK GDPR requirements, including the use of appropriate safeguards such as Standard Contractual Clauses or Binding Corporate Rules. The autonomous nature of agentic AI heightens the need for robust controls and clear contractual arrangements with third-party service providers.

d) Ethical deployment of agentic AI

Beyond legal compliance, ethical considerations are increasingly central to the responsible deployment of agentic AI. Organisations should consider establishing ethics committees or advisory boards to oversee the development and use of agentic AI, ensuring that systems are aligned with broader societal values and organisational principles. Transparent engagement with stakeholders, including customers and employees, can help build trust and mitigate reputational risks.

e) Prepare for future regulatory developments

The regulatory landscape for agentic AI is rapidly evolving, both in the UK and internationally. Organisations should monitor developments such as the EU AI Act, potential updates to the UK’s data protection regime and sector-specific guidance. Proactive engagement with regulators and participation in industry forums can help organisations anticipate and adapt to new requirements, ensuring long-term compliance and competitive advantage.

How Simmons & Simmons can help

We are already advising clients across the financial services, TMT and other sectors on the deployment of agentic AI. As this fast-developing and exciting area continues to evolve, we would be delighted to discuss any questions or concerns you may have about agentic AI and data protection compliance.

Please do not hesitate to reach out to your usual Simmons & Simmons contact or the contacts provided on this page.