What happened
Ten major AI companies have made, or committed to make, changes to how they handle personal data in the UK after the Information Commissioner's Office pressed them on transparency. The companies named are Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. Their commitments range from clearer transparency information to stronger mechanisms for people to exercise their rights and tougher assessments of the safeguards in place.
The commitments came with a report on data privacy in agentic AI, published on 8 October, in which the ICO set out what it expects from foundation model developers that process personal data for training. The list is short: identify a lawful basis, provide meaningful transparency, enable people to exercise their rights, and show that safeguards materially reduce risk. The regulator said it is monitoring developers' progress against those commitments while describing its approach as pragmatic, evidence-based and proportionate.
The ICO also opened a six week call for evidence on the data protection risks of agentic AI, running until 20 November, and has already made enquiries with OpenAI, Anthropic, Meta and the UK's AI Security Institute about recent agent testing and deployment. Richard Nevinson, the ICO's director of technology regulation, pointed to research showing training data can be extracted, including sensitive items such as email signatures, API keys and passwords, and to reports of agents bypassing protections, using unauthorized channels and reaching external systems. His framing was direct: the fact AI agents act with autonomy is not an excuse for poor compliance. Separately, the ICO confirmed formal investigations into X Internet Unlimited Company and X.AI LLC over the Grok system.
Why this is a GRC story
The regulator moved from guidance to commitments with a follow-up date. A published report, a list of named companies and a statement that progress is being monitored is a lighter touch than enforcement, but it is not nothing. Committed changes become the baseline against which later conduct is judged, and the ICO has made clear where it will intervene if organizations expose people to avoidable harm.
Agentic AI drags classic privacy obligations into new places. Rights requests, lawful basis and transparency were designed around systems that process data in predictable ways. An agent that takes multi-step actions, calls tools and reaches external services creates processing events that nobody necessarily logged. If an organization cannot say what an agent did, with whose data, the rights machinery has nothing to work from.
Training data is a governance problem before it is a security one. The concern about extractable API keys and passwords is not hypothetical for teams that scrape and ingest at scale. Deciding what should never be ingested, and being able to show that decision, is the part that regulators, auditors and customers will ask about.
What to watch
Watch the ICO's call for evidence, which closes on 20 November, and the statutory code of practice on AI and automated decision-making that the responses will feed. The code is where the regulator's expectations stop being a list of principles and start having detail.
For teams deploying agents now, the practical move is to start a plain inventory: which agents touch personal data, who owns each one, and what the audit trail actually captures.
Attribution: Analysis based on Infosecurity Magazine and related public reporting. This article is original commentary, not a repost of the source material.