A woman in a patterned dress stands against a clear blue sky, holding a remote controller and watching the small drone she is flying above her

The ICO said on 8 October 2026 that ten AI developers have made or committed to data protection changes after its review of their foundation models, and on the same day it opened a six week call for evidence on AI agents. The developers are Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. Between them they build the models behind almost every AI tool a law firm is offered, from the drafting assistant inside a practice management system to the chatbot a trainee opens on a phone.

The changes the ICO lists are clearer transparency information, stronger mechanisms for people to exercise their rights and tougher assessments of safeguards, and the regulator says it is monitoring progress against the commitments. An eleventh developer, xAI, was left out of the programme after the ICO opened a formal investigation into its Grok system. The report behind the announcement sets out the ICO’s positions on the lawful use of special category data in training and on whether a trained model can itself contain personal data. It also acknowledges that current training practices create technical challenges for compliance with UK data protection law, which the ICO says it is raising with Government.

What the commitments mean for a firm

A law firm is not party to any of this, and none of the ten has promised anything to you. What has changed is the information you can ask for. A supplier whose product runs on one of those models should be able to show you the developer’s transparency information and say what it tells users about training data and personal data. It should also explain how a subject access request reaching your firm would be answered at the model layer. Put that question to the supplier of every tool that touches client material, and record the answer beside the contract. For the consumer tools your staff use without telling you, the question is simpler, because those tools sit outside any contract your firm holds.

The agents, and the evidence the ICO wants

The second half of the announcement looks forward. Foundation models now underpin agents that complete tasks and use tools with limited human oversight, and the ICO says the data protection risks evolve as those systems gain autonomy. Its call for evidence asks developers, deployers and experts how they manage those risks across security, transparency, accountability, automated decision making, fairness and lawful data use. It closes on 20 November 2026 and will feed future guidance and the statutory code of practice on AI and automated decision making that sits in the ICO timetable.

The ICO also disclosed that it has made enquiries of OpenAI, Anthropic, Meta and the AI Security Institute after reports that, in testing, some agents bypassed protections, used unauthorised communication channels and reached external systems. Richard Nevinson, the ICO’s director of technology regulation, said that the fact AI agents act with autonomy is not an excuse for poor compliance. For a firm that is being sold an agent, the line to hold is that your firm is the deployer, and a deployer answers for what the agent does with client data. The NCSC’s seven controls remain the questions to put to the seller.

What to do before 20 November

List every tool the firm uses that runs on one of the ten models, and ask each supplier for the developer’s transparency information. Decide whether the firm has anything to say to the ICO about agents, because a small firm’s account of what it needs is evidence the regulator says it lacks, and the call is open to any organisation. And if an agent is already on your desk, write down who in the firm approves what it may do with client data before it runs, since that is the question the ICO is now putting to the developers themselves.

The ICO’s announcement of 8 October 2026, ICO secures changes from leading AI developers as scrutiny extends to AI agents, is on ico.org.uk, with the foundation model report and the call for evidence linked from it.

If your firm wants a plain answer on which of its tools run on which models, and what to ask each supplier, start with a conversation.