What ICO AI Data Protection Commitments Mean for AI Deployers

The UK Information Commissioner’s Office said on October 8, 2026 that 10 foundation model developers have made, or committed to make, changes to how they handle personal data, the first formal set of ICO AI data protection commitments covering the largest model builders.

The developers are Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. The ICO published the results in a report titled Building trust and transparency into generative AI development: our work to create regulatory certainty, alongside a news announcement.

For compliance and privacy teams at firms that build on these models, the report sets out what the regulator expects upstream. It also shows where the ICO has not yet spoken, including the duties of companies that deploy models rather than train them.

What the ICO AI data protection commitments cover

The ICO’s position is that developers training models on personal data must have a lawful basis for doing so. They must also give people meaningful transparency, let them exercise their rights, and keep “safeguards in place to materially reduce risk.”

The regulator said the changes secured so far include clearer transparency information, stronger mechanisms for people to exercise their rights and tougher assessments of safeguards. It did not say which developer made which change.

The report also sets out ICO positions on two difficult questions. One is how special category data, such as health or biometric information, can be used lawfully in training. The other is whether a foundation model can itself contain personal data. Where a model does, the ICO says the bar is “even higher.”

The regulator cited research showing that training data can be extracted from models, including material taken from the internet such as email signatures, “API keys and passwords.”

Are the commitments voluntary or enforceable?

The announcement describes changes that developers “made or committed to make” after ICO scrutiny. It does not publish the commitments as legal undertakings, and it sets no deadlines or penalties for the 10 developers. The ICO said it is monitoring progress against the commitments.

The legal duties themselves are not new. The ICO said the practices it is concerned about fall under existing law, mainly the UK General Data Protection Regulation (UK GDPR), which already requires a lawful basis, transparency and respect for individual rights. The ICO can use its enforcement powers under that law where those duties are not met.

The regulator said it will intervene where organizations expose people to avoidable harm or proceed without adequate safeguards. It described its approach as pragmatic, evidence-based and proportionate, and said it will work with developers that engage constructively.

X.AI faces a formal ICO investigation

X.AI is not among the 10. The ICO has opened formal investigations into X Internet Unlimited Company and X.AI LLC over the Grok AI system’s potential to produce harmful sexualized image and video content. It paused its engagement with X.AI after opening the inquiry, and the investigation is ongoing.

The split shows how the ICO is using its supervision program. Developers that engage get dialogue and monitoring. A developer under formal investigation does not.

The foundation model supervision program was created in 2025 under the ICO’s AI and biometrics strategy. It covered 11 priority developers, chosen by likelihood of non-compliance, UK market share and use of higher-risk training datasets. The ICO’s March 2026 strategy update described that engagement and promised further policy positions to give developers more certainty.

What the commitments mean for teams deploying models

The report is addressed to developers. It does not discuss the duties of deployers, and it does not mention data protection impact assessments. Firms using these models in products or internal workflows therefore have no new ICO checklist to work from.

What changes is the evidence available upstream. Clearer transparency information from developers gives deployers more material for their own privacy notices, vendor due diligence and records of processing. Teams running third-party risk monitoring can compare vendor documentation against the ICO’s stated expectations on lawful basis, individual rights and safeguards.

The ICO also flagged that current training practices create technical challenges for UK data protection law and for data protection by design. It said it is raising those challenges with the government.

Deployers remain controllers of their own processing. A developer’s commitment to the ICO does not change a deploying firm’s obligations under the UK GDPR, which stay the same whichever model sits underneath. For a wider view of the compliance workload, see this guide to GDPR compliance strategies.

Agentic AI is the next ICO focus

Alongside the report, the ICO opened a call for evidence on agentic AI, meaning systems that act with a degree of autonomy. It runs for six weeks and closes on November 20, 2026. The ICO is seeking views from developers, deployers and experts on security, transparency, accountability, automated decision-making, fairness and lawful data use.

The responses will inform future guidance and the forthcoming statutory code of practice on AI and automated decision-making. That code is the document most likely to bind deployers directly.

The ICO has also made enquiries with OpenAI, Anthropic, Meta and the UK AI Security Institute about recent tests and deployments of AI agents. It has asked about the risk assessments and safeguards in place. The enquiries are ongoing and the ICO has given no outcome.

Richard Nevinson, Director of Technology Regulation at the ICO, said: “AI has huge potential to benefit our society, but that depends on trust and transparency.” He added that “the fact AI agents act with autonomy is not an excuse for poor compliance.”

How the rules fit together

The Data (Use and Access) Act 2025 amended the UK’s data protection framework and restructured the regulator. It moved the ICO from a corporation sole, headed by one commissioner, to a collective Information Commission Board. The ICO’s report treats the UK GDPR as the operative standard for training data.

In the EU, Regulation (EU) 2024/1689, the AI Act, sets separate obligations for providers of general-purpose AI models, which have applied since August 2, 2025. Those duties cover areas such as technical documentation and training content summaries. Nine of the 10 developers named by the ICO operate in both markets, so a firm deploying their models in the UK and the EU faces two rulebooks that cover overlapping ground. Readers can find background in this overview of the EU AI Act.

The ICO’s own role as the UK data protection regulator means its positions on special category data and on personal data inside models will shape how UK courts and organizations read the UK GDPR, even without new legislation.

What comes next

The ICO has set no date for reviewing the 10 developers’ progress. The next fixed date is November 20, 2026, when the agentic AI call for evidence closes. The ICO’s investigation into X.AI over Grok continues.