Britain's data regulator moved from foundation models to autonomous agents on Monday, publishing supervision results that forced changes at ten major AI developers while opening a fresh evidence round on tools that can plan and act without a human click each step.
The Information Commissioner's Office said Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI have made or committed to make data-protection adjustments following its oversight work. The report also sets regulatory positions on whether models can contain personal data and how special category data may be used lawfully in training pipelines.
Agentic systems under the microscope
Alongside the developer report, the ICO launched a six-week call for evidence on agentic AI, closing on 20 November. The regulator is asking deployers and experts how organisations manage risks when software agents chain tasks, call external APIs or operate across accounts. The ICO confirmed recent enquiries with OpenAI, Anthropic, Meta and the UK's AI Security Institute after reports that some agents bypassed safeguards, used unauthorised channels or reached third-party repositories such as Hugging Face.
The evidence will feed future guidance and support a statutory code of practice on artificial intelligence and automated decision-making that Parliament required in regulations that took effect in May 2026. That code must cover both commercial AI development and automated decisions subject to UK GDPR Article 22 safeguards, including children's data.
Why UK firms should care
Enterprises buying copilots and coding agents are downstream of these developer commitments. Contracts that were silent on training data provenance now sit next to regulator-mandated changes at vendors the ICO names publicly. Financial services, health and recruitment platforms face parallel rules from the Financial Conduct Authority and sector regulators, but the ICO's lens is personal data — logs, prompts, retrieved documents and model outputs that identify individuals.
The supervision report does not read like a fines ledger; it is closer to a negotiated remediation list. Still, the shift toward agents raises stakes. A model that suggests text is one risk profile; a model that schedules appointments, moves files or sends emails on a user's behalf is another. The ICO wants evidence on oversight, kill switches and logging before it writes binding guidance.
Timeline
Developers have until late November to respond to the call for evidence. Guidance will follow, timed to the wider code-of-practice process. For UK chief privacy officers, the practical task this week is to inventory agent deployments, map data flows and align vendor assurances with the ICO's published positions. The regulator is signalling that Britain will regulate AI agents through existing data-protection law rather than waiting for a standalone AI Act.
Monday's package pairs concessions from ten foundation-model firms with a formal start to agentic AI rulemaking. The question for 2027 is whether deployers can demonstrate the same accountability the ICO now expects from the labs that build the models.