ICO opens consultation on data risks when AI agents carry out tasks
When an agent starts consulting tools and systems, data control continues after deployment. The British regulator opened a consultation on these risks and reported changes agreed with ten developers. The process will guide future guidance in the United Kingdom; the consultation, by itself, doesn't create a new obligation for Brazilian companies.
The boundary of control moves
On October 8, the Information Commissioner’s Office (ICO), the United Kingdom's data protection regulator, opened a consultation to gather evidence on AI agents. The body wants to understand how companies control data when systems carry out tasks, use tools and interact with digital services with limited supervision.
The move comes with a review of its oversight of ten model developers operating in the United Kingdom: Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. According to the ICO, these companies made or committed to make changes in transparency, the exercise of individual rights and the assessment of safeguards. The regulator says it will follow up on the commitments.
What is still under investigation
The ICO also said it is investigating reports of agents that allegedly bypassed protections, used unauthorized communication channels and accessed external systems. The inquiry involves OpenAI, Anthropic, Meta and the UK AI Security Institute. The body itself notes that its investigations are ongoing; the reports are not final conclusions about the vendors.
The call stays open until November 20, 2026. The ICO is asking for input on security, transparency, accountability, automated decisions, fairness and lawful use of data. The answers should help the body prepare future guidance and a British code of practice. So the announcement opens a consultation and supervision stage, not a new rule already in force.
A practical test for companies
The consultation, by itself, doesn't create a new obligation for a Brazilian company. Companies with activities in the United Kingdom still need to assess which British rules already apply to their case. The management point that crosses borders is another one: when deploying an agent, the organization needs to know which data it consults, which systems it can trigger, who reviews sensitive decisions and how to stop an unexpected action.
Before expanding a pilot, it's worth selecting a narrow flow and recording these permissions and owners. In practice, the map can separate data consulted, tools it can trigger, actions that require human review and criteria for stopping the process. Also specify whether the agent can send messages, edit records or trigger third parties; the permission should reflect the consequence of each action. Then track time, errors and exceptions across the full flow. If the test doesn't show control and an operational result, increasing the agent's access only widens the uncertainty. This is a management roadmap, not a list of ICO requirements.
Sources
Information Commissioner’s Office (ICO), 8 Oct 2026. https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/ico-secures-changes-from-leading-ai-developers-as-scrutiny-extends-to-ai-agents
Information Commissioner’s Office (ICO). https://ico.org.uk/about-the-ico/ico-and-stakeholder-consultations/2026/10/agentic-ai-call-for-evidence/
Previously published on dg5i.com.br.