Accompanying the report is a blog post from the ICO media centre, playfully titled “AI’ll get that!”, which paints a vivid picture of how these “agentic” systems could soon handle everything from grocery shopping to booking holidays. But beneath the excitement lies a serious message: without robust data protection, these powerful tools could erode privacy in ways we’re only beginning to understand.
What Exactly Is Agentic AI?
The ICO defines agentic AI straightforwardly but with important nuance: it’s the combination of generative AI capabilities (like those in large language models) with additional tools and new interaction methods that allow systems to operate more autonomously in the world.
Unlike today’s chatbots, which respond to prompts and stop there, agentic systems can:
- Maintain context over long periods
- Use natural language to plan and reason
- Access external tools (browsers, APIs, databases)
- Execute multi-step tasks toward open-ended goals
Think of it as giving AI “agency”—the ability to act independently to achieve objectives we set (or that it infers).
The report emphasizes that agentic AI builds on generative foundations but shifts the paradigm. Where generative AI excels at creating content, agentic AI focuses on action: researching options, making transactions, coordinating schedules, or even negotiating on your behalf.
Key Characteristics Highlighted by the ICO
The report lists several defining traits:
- Automation of complex, open-ended tasks: Moving beyond rule-based automation to handling ambiguity and nuance.
- Increased automated decision-making: Agents may decide independently within delegated authority.
- Broad purposes: Goals are often set vaguely (“manage my finances”) rather than narrowly.
- Access to diverse data sources: Agents pull from emails, calendars, bank accounts, browsing history, and more.
- Potential for inference: Systems might deduce sensitive information (health conditions, political views) from patterns.
- Cybersecurity vulnerabilities: New attack surfaces from tool use and external interactions.
These characteristics make agentic AI powerful but also harder to control and oversee.
The Promise: Transformational Applications Across Life
The ICO is clear that agentic AI isn’t just hype—it’s already emerging in prototypes and early products. Applications span multiple domains:
Personal Assistants and Consumer Space
The blog post focuses heavily on “agentic commerce,” envisioning personal shopping “AI-gents” that proactively manage purchases. Examples include:
- Monitoring your calendar and habits to anticipate needs (e.g., buying running shoes before a marathon training plan starts).
- Checking bank balances to ensure purchases fit budgets.
- Timing buys for sales events like post-Christmas clearances.
- Negotiating prices directly with retailers.
- Presenting financing options for approval.
Broader consumer uses could extend to travel booking, household finance management, or even meal planning with automatic grocery orders.
Workplace and Enterprise
In professional settings, agents might handle research, coding, project planning, email triage, or customer interactions with greater autonomy.
Government and Public Services
Potential for streamlining benefits applications, tax filing, or permitting processes—though with heightened scrutiny on fairness and accountability.
Healthcare and Cybersecurity
Agents could monitor patient data for anomalies or defend networks by responding to threats in real time.
The ICO acknowledges genuine benefits: efficiency gains, accessibility for those with disabilities, and innovation in privacy-enhancing tools themselves (e.g., agents that audit data usage for compliance).
The Privacy Risks: Why Agentic AI Raises New Concerns
While sharing some risks with generative AI (like inaccurate outputs or bias), agentic systems introduce novel challenges due to their action-oriented nature.
Novel Risks Identified in the Report
The ICO outlines several interconnected issues:
- Broad and unclear purposes: When users instruct an agent to “handle my shopping,” the purpose for processing personal data becomes vague, potentially violating data minimization principles under UK GDPR.
- Excessive data processing: Agents may access far more information than strictly necessary—pulling in entire email histories or financial records “just in case.”
- Inference of special category data: From spending patterns, an agent might deduce health conditions (e.g., frequent pharmacy purchases) or religious beliefs, triggering stricter GDPR rules.
- Transparency and rights exercise: Increased complexity makes it harder for individuals to understand decisions or exercise rights like access, rectification, or erasure.
- Controller-processor uncertainty: In complex supply chains (foundation model + tool integrations + deployer), determining who is responsible for compliance becomes murky.
- Cybersecurity threats: Agents with access to sensitive tools create new vectors for attacks, such as prompt injection or data exfiltration.
- Data concentration: Personal assistant agents could amass vast profiles in one place, amplifying risks if breached.
Poor design exacerbates these: agents connected to unnecessary databases, lacking monitoring, or without “stop” mechanisms.
Real-World Scenarios from the Blog
The shopping agent example illustrates risks vividly. To “get that” deal, your AI-gent needs deep access to:
- Financial data
- Location history
- Calendar
- Preferences
- Browsing behavior
One misjudgment or overreach, and sensitive insights are processed without explicit consent.
William Malcolm, ICO’s Executive Director of Regulatory Risk and Innovation, warns: while benefits are transformational, “strong data protection foundations are required to build trust and scale safe AI adoption.”
Regulatory Landscape: UK GDPR in an Agentic World
Organisations remain fully accountable under UK GDPR and the Data Protection Act 2018, regardless of how “autonomous” the AI seems.
Key implications:
- Lawful basis: Reliance on consent or legitimate interests becomes trickier with broad, evolving processing.
- Data Protection Impact Assessments (DPIAs): Essential for high-risk deployments.
- Automated decision-making (Article 22): If agents make legal or significant decisions without human oversight, special rules apply.
- Accountability: Documentation of purposes, data flows, and risk mitigations is critical.
The report notes upcoming changes under the Data (Use and Access) Act, with consultations planned for 2026 on automated decision-making.
ICO Recommendations: Building Privacy In
The ICO urges caution but encourages innovation. Practical advice includes:
For Developers and Deployers
- Define clear, narrow purposes for processing.
- Limit data access to what’s strictly necessary.
- Implement governance: monitoring, human oversight, emergency stops, and sharing controls.
- Use privacy-enhancing features (e.g., on-device processing where possible).
Innovation Opportunities
The report highlights “privacy-positive” potential:
- Agents that help individuals manage their own data rights.
- Information governance agents that audit compliance.
- Benchmarking tools for evaluating agentic systems’ privacy impact.
Organizations are encouraged to use ICO services like the Regulatory Sandbox for testing innovative uses.
Looking Ahead: Four Scenarios for the Next 2-5 Years
The report uses horizon-scanning with four plausible futures to stress-test implications:
While specific scenario details aren’t fully elaborated in summaries, they explore variables like adoption speed, capability breakthroughs, and regulatory responses—ranging from widespread consumer agents to more constrained enterprise use.
The ICO’s message: outcomes depend on choices made now.
The ICO’s Role and Next Steps
As the UK’s data protection authority, the ICO positions itself as both guardian and enabler:
- Monitoring developments throughout 2026.
- Collaborating with developers via workshops and guidance.
- Partnering internationally (e.g., G7 data protection authorities).
- Updating guidance on generative and agentic AI.
The public is invited to report concerns, reinforcing democratic oversight.
Agentic AI Forces Organizations to Balance Transformation with Trust
Agentic AI stands to reshape daily life profoundly—freeing us from mundane tasks and opening new possibilities. Yet as the ICO’s Tech Futures report makes clear, this future hinges on getting privacy right from the start.
The playful blog title “AI’ll get that!” captures the convenience, but the underlying warning is serious: without purposeful design, strong governance, and clear accountability, these agents could “get” far more personal information than we intend.
For businesses rushing to deploy agentic systems, the message is straightforward: embed data protection by design, engage early with regulators, and view privacy not as a barrier but as the foundation for sustainable innovation.
As we step into 2026, the ICO’s work serves as an essential guide—helping ensure that the age of autonomous AI agents is one of empowerment, not erosion of personal privacy.
