The UK’s data protection regulator has secured commitments from 10 major artificial intelligence developers to improve safeguards around personal information, as it investigates reported incidents involving AI agents and seeks evidence on the risks posed by increasingly autonomous systems.
The Information Commissioner’s Office (ICO) said Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI had made, or committed to make, changes following regulatory scrutiny of their foundation models.
The improvements include clearer information about how personal data is used, stronger mechanisms for individuals to exercise their data protection rights, and more robust assessments of safeguards. The ICO said it would monitor progress against the developers’ commitments.
The announcement marks a shift in the regulator’s focus from the models underpinning generative AI tools to the agents increasingly built on top of them, which can use tools, interact with websites and carry out tasks with limited human oversight.
The ICO has launched a six-week call for evidence on the data protection implications of agentic AI, seeking responses from developers, deployers and other experts by 20 November. The findings will inform future regulatory guidance and a statutory code of practice on AI and automated decision-making.
The regulator has also confirmed enquiries involving OpenAI, Anthropic, Meta and the UK’s AI Security Institute following reports of agentic AI testing and deployment earlier this year.
According to the ICO, some agents reportedly bypassed safeguards, used unauthorised communication channels and accessed external systems, including Hugging Face. The regulator is seeking to establish what risk assessments and protections were in place at the time. Its enquiries remain ongoing.
The incidents raise questions about how organisations can maintain effective oversight when AI systems are able to take actions with limited human intervention, potentially creating new risks around access to personal information, accountability and compliance.
“AI has huge potential to benefit our society, but that depends on trust and transparency,” said Richard Nevinson, Director of Technology Regulation at the ICO. “Our message is clear: the fact AI agents act with autonomy is not an excuse for poor compliance.”
The call for evidence will examine issues including security, transparency, accountability, fairness, lawful data use and automated decision-making. The ICO said the responses would help it clarify how organisations should manage the risks while allowing responsible development of the technology.
The ICO’s latest report also sets out its regulatory position on the use of personal information to train foundation models, which underpin many chatbots and other AI applications.
The regulator has highlighted questions over the lawful use of special category data, which includes information such as health records and political opinions, and whether foundation models themselves can contain personal data.
It acknowledged that current foundation model training practices present technical challenges for compliance with UK data protection law and the principle of data protection by design. It said addressing these issues would require continued collaboration between developers, regulators and government.
The ICO established a dedicated foundation model supervision programme in 2025, covering 11 priority developers. Its engagement with X.AI was paused after the regulator opened a formal investigation into its Grok AI system, which remains ongoing.
Alongside its work on foundation models and agents, the regulator is examining the growing personalisation of consumer-facing AI services, including general-purpose chatbots and AI companions. It is conducting public research into potential concerns and engaging with companies on transparency and privacy.
As AI systems move beyond generating responses to taking actions, the regulator faces a more complex task: ensuring that data protection safeguards remain effective not only during training and deployment, but also as systems operate with increasing autonomy.
There’s plenty of other editorial on our sister site, Electronic Specifier! Or you can always join in the conversation by commenting below or visiting our LinkedIn page.
