The Most Dangerous Thing Generic AI Is Selling to Social Housing

By Phil Shelton MBA CIHM, Chief Executive, HousingAI

It’s become fairly common for housing associations to have a director of transformation and innovation. Most conferences have a panel where someone explains that the sector cannot afford to be left behind.

None of that is wrong. But there is a version of AI adoption that looks like progress and isn't. Let’s call it the illusion of progress.

Between the enthusiasm and the evidence

Starbucks spent years and significant capital building an AI-powered ordering system before pulling it. The technology worked in demos but couldn't handle the operational reality of thousands of stores, hundreds of customisations, and staff who needed to do several other things at once. A Waymo autonomous vehicle in San Francisco was pursued by police after it drove through a construction zone on a motorway and couldn't be stopped - because there was no obvious way to intervene in a system designed to operate without human input.

The housing sector is not immune. The pilot that runs for six months and produces a great dashboard, but there is no behaviour change. The Copilot rollout that prompts staff mid-document - would you like me to write this for you? - and becomes the source of guidance nobody has verified. The board report generated from an Excel spreadsheet by a tool that confidently missed three of the most material figures.

Enough organisations have tried generic AI to know it is not straightforward. But that gap - between the enthusiasm and the evidence - is where the most expensive mistakes are made.

The "silver bullet" problem

Inside Housing's recent review of social landlords' annual reports found genuine ambition - Platform Housing Group using machine learning to identify 556 "silent" tenants for welfare checks; Peabody testing AI thermal imaging drones for retrofit data. But the overarching mood in those reports was caution, and for good reason. The risks flagged most consistently were data privacy, cybersecurity, and the danger of an AI-driven "computer says no" culture removing human judgement from decisions that affect vulnerable people.

That anxiety is well-founded. And it sits awkwardly alongside the pressure to be doing something.

The result is a pattern the sector has seen before with other technologies. A wave of pilots, a lot of noise about transformation, and then a slowdown as organisations hit governance questions. How is the data governed? What happens if the output is wrong? Who is accountable? What does the audit trail look like?

Generic AI - Copilot, ChatGPT, Gemini - is not designed to answer those questions for a regulated sector. It is designed to be broadly useful across the widest possible range of use cases. That generality is a feature for most purposes. In housing compliance, it is a liability.

Housing providers are dealing with increasing regulatory complexity, while expectations around evidence and accountability continue to rise. HousingAI has been built to give teams a practical way to navigate that. It brings together regulation, guidance, and best practice into something people can actually use, whether that's reviewing a policy, preparing for inspection, or sense-checking a decision.

The regulatory ratchet

UK MPs are currently backing an amendment to the Cyber Security and Resilience Bill that would give the government a literal kill switch to shut down data centres or AI systems posing a threat to national security or essential services. It sounds like science fiction, but it is currently being debated in Parliament.

For housing providers, the direction of travel is clear. Deploying AI that relies on opaque, open-internet models, with no clear data sovereignty, no audit trail, and no legal validation of outputs, is no longer just a technical risk. It is a regulatory one. The organisations that treat "we use Copilot" as an AI strategy are building on sand.

Stoke-on-Trent City Council's deployment of AI to identify early warning signs in tenant payment behaviour points in a better direction. The system is designed not to automate decisions but to obtain insight early enough for officers to intervene. Human judgement prevails, and the AI handles the pattern recognition at a scale that officers would struggle to handle. That is a good division of labour.

Secure and sector-specific

The organisations that will get the most from AI in housing are not the ones that move fastest. They are the ones who ask the right questions before they deploy:

• What is this tool trained on?

• Where does the data go?

• What does the output cite?

• What happens if it is wrong?

The focus has always been on making this work in the reality of housing, so it's something teams can rely on when it matters most.

The silver bullet does not exist. But a secure, sector-specific tool - built on validated housing sources, with Anthony Collins providing legal commentary on what the changes mean - is not a distant ambition. It delivers consistent, cited guidance at the point of decision and creates an audit trail that holds up when a regulator or ombudsman comes looking. It is available now.

The gap between AI enthusiasm and AI infrastructure is a governance question. Housing has the regulatory pressure, the sector expertise, and now the tools to close it. The only real variable is whether organisations are willing to ask what they are actually deploying and why.