AI in supported housing: what the regulator expects and the guardrails to set
How the Regulator of Social Housing treats AI, where technology helps in exempt accommodation, and the data protection and oversight guardrails boards need.
Quick answer
The Regulator of Social Housing has no separate AI rules. It judges AI through its existing standards, so if a tool gets a repair, a complaint or a tenancy decision wrong, that is the landlord's failure, with no software defence. In supported housing, add a DPIA before any tool touches resident data, a named person behind every decision about a resident, and a check that no resident is shut out.
The short version
- The RSH says it will apply its existing standards to AI and act only where the outcomes they require are compromised. Responsibility stays with the landlord.
- The most useful technology in exempt accommodation joins support records to repairs and property compliance. Resident-facing chatbots come a long way down the list.
- Support records hold special category data. Do a DPIA first, and keep resident details out of consumer AI tools.
- UK GDPR still restricts significant decisions based solely on automated processing of special category data, so a person must decide.
- Automation layered over messy records and unclear managing agent agreements makes failures happen faster.
In this guide
Landlords are moving quickly on AI, from repairs triage to drafting letters and analysing data. The regulator has now said how it will judge that, and the answer is simple: it looks at what happened to the tenant, not at the tool.
What the regulator has said
In a comment piece for Inside Housing on 13 October 2026, Kate Dodsworth of the Regulator of Social Housing set out its approach [1]. The main points:
- No separate AI rules. The regulator will apply its existing standards and will intervene only where the outcomes those standards require are compromised.
- No software defence. Responsibility sits with the landlord. If an AI tool causes a failure, it is assessed as a failure against the standards.
- A person behind every decision. AI can analyse and advise, but decisions that affect people's homes need someone accountable.
- Framework first. Landlords should have proper oversight in place before significant adoption, and be open with tenants about when AI is in use.
- Known risks. Inaccurate data, model errors, bias, cyber security, digital exclusion and accessibility.
The regulator is also changing how it works. Its Corporate Plan 2026-2029, published on 4 November 2026, says it will seek to increase its use of automated data analysis and emerging technologies such as AI, alongside a new digital, data and technology strategy [2]. The same plan repeats its concern about the risks of lease-based specialised supported housing [2].
Where technology earns its place in exempt accommodation
The biggest problems in supported housing records are rarely solved by a chatbot. They sit in the gap between what support staff hear and what the property side does about it.
- Hazards in support notes reaching repairs. A resident tells a support worker the heating has gone or there is water coming through a light fitting. A form or keyword check that turns that into a repair job the same day, with a reference number, closes the most common gap we see. From 30 November 2026, Awaab's Law Phase 2 adds excess cold, electrical hazards and other hazards for registered providers' tenancies, and the clock can start when a managing agent becomes aware [7]. It does not apply to homes occupied under a licence. Our Awaab's Law Phase 2 update explains the test.
- Joining up systems across organisations. Support providers, landlords and managing agents often run separate systems. Matching a repair logged by support staff to the agent's job and the landlord's completion record shows who owns each repair, and where it stalled.
- Checking interests before someone else does. Matching board members, staff and managing agents against Companies House records can show links to related companies that nobody declared. The RSH inquiry into Ash-Shahada is about conflicts of interest and third-party payments, so this is a live governance question.
In each case the tool raises a flag. A person still acts on it, and the record shows they did.
The guardrails to set before anything goes live
| Do | Don't |
|---|---|
| Complete a DPIA before any tool touches resident data | Paste resident names, health details or safeguarding disclosures into consumer AI tools |
| Keep a named, trained person behind every decision about a tenancy, support level or safeguarding | Let a score decide who gets help first without a person checking it |
| Check that residents who report by phone, in person or through staff are not pushed down the queue | Assume a tool is fair because the supplier says so |
| Fix the records and the managing agent agreement first | Layer automation over inconsistent spreadsheets |
| Tell residents when AI is used and what a person still decides | Hide AI use in a privacy notice nobody reads |
Data protection
Support records hold health, substance use, offending and safeguarding information. Most of it is special category data under UK GDPR, which needs a lawful basis and an extra condition for processing [6]. The ICO expects a DPIA where processing is likely to be high risk, and new technology applied to sensitive data about vulnerable people is the clearest case [5].
Consumer AI accounts are the practical risk. If a worker pastes a resident's details into a tool your organisation has no contract with, you cannot say where that data went or how long it is kept. Use only tools covered by a data processing agreement, set out in the DPIA, with staff told plainly what can and cannot go in.
Automated decisions
The Data (Use and Access) Act 2025 rewrote the UK GDPR rules on automated decision-making from 5 February 2026 [4]. The rules are looser for ordinary personal data, but a significant decision based on special category data still cannot be made by entirely automated means unless the person gives explicit consent or a legal condition applies [3]. In supported housing, decisions about support levels, evictions and safeguarding will usually be both significant and based on that data. Keep a person in the decision and record that they made it.
Fairness and access
Many residents in exempt accommodation have no smartphone, limited data, low literacy or a disability that makes online forms hard. A system that ranks repairs or contact by digital channel can disadvantage them. That can be indirect discrimination under the Equality Act 2010 [8], and for registered providers it touches the fairness and complaints parts of the consumer standards [1]. Test any prioritisation against the residents who report in person or through staff.
Getting ready for licensing
Licensing is likely to test governance and records more than technology. Expect to be asked where a piece of evidence came from, who wrote it and who checked it. Before you add tools, make sure you can answer three questions for every system that holds resident data:
- What data goes in, and who can see it?
- Which decisions does it inform, and who makes them?
- How would you show a council officer that a flag was acted on?
If the answers are unclear today, automation will not make them clearer. Our guide for managing agents and proptech teams lists the data points councils ask for, and the draft licensing regulations update covers what is proposed.
For what this looks like inside a support file, Support Audit has a guide to auditing digital support records, including how to spot AI-padded session notes.
Your next step
List every tool that touches resident data, who owns it and whether a DPIA exists. Gaps on that list are your first fixes.
Questions people ask
Does the Regulator of Social Housing have rules for AI?
No separate rules. In October 2026 the regulator said it will apply its existing standards, including those on complaints, fairness, governance and transparency, and will intervene only where the outcomes those standards require are compromised. An AI failure is assessed as a standards failure.
Can support workers use ChatGPT or other AI tools to write support notes?
Not with resident details in them through a consumer account. Names, health information and safeguarding disclosures are personal data, much of it special category data, and putting them into a tool without a data processing agreement, a DPIA and a lawful basis puts the provider in breach of UK GDPR. Even with an approved tool, the worker should write the note from what happened and check every word.
Do we need a DPIA before using AI in supported housing?
In almost every case, yes. The ICO expects a DPIA where processing is likely to be high risk, and using new technology on special category data about vulnerable people is the clearest example. Do it before the tool goes live, not after.
Do these rules apply if we are not a registered provider?
The RSH standards apply to registered providers. Data protection and equality law apply to everyone, and councils will judge records and governance when supported housing licensing starts. A non-registered provider has no reason to set a lower bar.
Can AI decide which repairs or residents to prioritise?
It can suggest and route. A person should own the decision and be able to explain it. A scoring system that ranks reports by how they were made can push residents who report in person, by phone or through a support worker down the queue, which is an equality risk as well as a safety one.
Sources
Primary legislation and official guidance come first. Where we rely on commentary, we say so.
- 1Kate Dodsworth: the regulator's approach to AI Inside Housing, 13 Oct 2026
- 2Regulator of Social Housing Corporate Plan 2026-2029 Regulator of Social Housing, 4 Nov 2026
- 3UK GDPR, Article 22B: restrictions on automated decision-making legislation.gov.uk
- 4Data (Use and Access) Act 2025, section 80 legislation.gov.uk
- 5When do we need to do a DPIA? Information Commissioner's Office
- 6Special category data Information Commissioner's Office
- 7Awaab's Law Phase 2: Guidance for social landlords MHCLG
- 8Equality Act 2010, section 19: indirect discrimination legislation.gov.uk
Regulation alerts
Get alerts when this changes
One short email when licensing, Housing Benefit or national standards change, with what it means for your service. About once a month.
Support Audit
Want a second pair of eyes on your schemes?
Support Audit reviews support files, Housing Benefit evidence, safeguarding and property standards against the incoming national standards, then gives you a plain-English action plan.
Read next
All guidesA guide for managing agents and proptech teams in supported housing
Who carries which duty in supported housing, the data councils and licensing will ask for, and how managing agents and proptech teams can build for it.
Updated Nov 2026
Licensing and the national supported housing standards
What the Supported Housing Regulatory Oversight Act does, what is in force, what the draft licensing regulations add and what to do before 10 December 2026.
Updated Nov 2026
Property standards and safety in supported housing
The property rules for supported housing: HHSRS, HMO licensing, fire safety, gas and electrical checks, smoke and CO alarms, Awaab's Law and Decent Homes.
Updated Oct 2026