The United Kingdom’s Information Commissioner’s Office (ICO) has secured data-protection improvements from ten major AI developers and opened a new call for evidence on the privacy risks posed by autonomous AI agents.
The regulator announced the developments on October 8, following scrutiny of foundation-model providers including Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI.
According to the ICO, the companies have either made or committed to making changes following regulatory engagement. The announcement did not mean that every concern had been resolved or that every company had made identical changes.
The regulator also confirmed that it is making enquiries into agentic AI testing and deployment earlier in 2026. Its new call for evidence seeks further information about how systems that can take actions on behalf of users may affect personal data and privacy.
Why AI agents create new privacy risks
Traditional chatbots generally respond to prompts. AI agents can go further by interacting with software, using tools, processing files and carrying out multi-step tasks with varying degrees of independence.
These capabilities can save time, but they can also expand the amount of personal information an AI system can access. An agent connected to email, calendars, customer databases or workplace applications may encounter sensitive information that is not necessary for its task.
The risks increase when an agent has broad permissions or can pass information between different services. A poorly configured system might expose data to an unintended recipient, retain information unnecessarily or take an action that the user did not fully anticipate.
The ICO’s latest initiative reflects the need to assess AI not only as a technology that generates text, but also as software that may act on information and interact with real systems.
Foundation models are used across thousands of products
Many AI applications depend on a relatively small number of foundation models. Those models are trained on large datasets and can be adapted for different uses, from general-purpose assistants to business software.
Because they sit beneath many applications, changes made by foundation-model developers can have consequences well beyond a single product. Better data-handling practices at the model level may improve the privacy protections available to companies and individual users further down the chain.
However, responsibility does not rest solely with model providers. Businesses that integrate AI into their services must also consider what information they supply, what permissions they grant and how they handle the outputs.
What businesses should review
Organisations using AI agents should start by checking which systems and data each agent can access. Permissions should be limited to what is necessary for the task, and sensitive actions may require explicit human approval.
Businesses should also examine how personal information is retained, whether users understand how their data is processed, and what happens when an agent makes a mistake. Testing should cover realistic scenarios, including attempts to manipulate the system or persuade it to disclose information.
Privacy notices and internal policies need to reflect actual behaviour, rather than simply stating that a company uses AI responsibly. Organisations should be able to explain what information enters an AI workflow, where it goes and who is accountable for the result.
A developing regulatory challenge
The ICO’s call for evidence is an opportunity for researchers, developers, businesses and members of the public to provide information about the risks and practical safeguards associated with agentic AI.
The process could help the regulator identify where existing privacy protections are sufficient and where additional guidance or enforcement may be needed. Its findings will be particularly relevant as AI agents become more integrated into workplace software and consumer services.
The broader lesson is that privacy protections must evolve alongside the capabilities of AI. A system that can access files, send messages or update records needs safeguards that go beyond checking whether its generated text is accurate.
Why it matters: The ICO’s action signals that regulators are increasingly examining what AI systems do with personal information—not just what they say. For companies deploying AI agents, limiting access and maintaining human oversight are becoming central parts of responsible implementation.
Source: UK Information Commissioner’s Office, October 8, 2026 — ICO secures changes from leading AI developers as scrutiny extends to AI agents.
Laat een reactie achter