AI and legal-content assurance

TenantSafe is in beta, with first customer delivery on 20 August 2026. This page describes how it works today. Where something is planned rather than built, it says so.

Where AI is used, and where it is not

There are two products and they use AI differently.

• TenantSafe uses a large language model for two things: guided diagnosis, which turns a resident's plain-English description into a categorised issue, and letter generation, which drafts a formal complaint letter with the statutory references that apply. Anthropic is the model provider for letter generation.
• Statutory deadline calculation is not AI. The Complaint Handling Code and Awaab's Law clocks are computed in code from recorded dates. A model is never asked what a deadline is, and cannot move one.
• ScoreView synthesises published Ombudsman determinations and regulatory judgements. Every briefing it produces is stamped with the record count and the corpus timestamp behind it, so a reader can see how much of the record the answer was drawn from and when that record was last taken.

What data reaches the model

The resident's own description of the problem is sent to the model provider. That is what makes the diagnosis and the letter possible, and there is no version of the product where it is not. A complaint narrative written by a resident can contain health information, details of household members, and the circumstances of the home, and you should assume it does.

We are not going to tell you that the narrative is stripped or redacted first, because it is not. If your information-governance position requires that special category data never leaves your estate in free text, say so early: that is a real constraint and it needs to be worked through before deployment, not discovered during it.

A data protection impact assessment covering this processing exists and we will share it under NDA with a prospective customer's IG team.

Getting the specifics for your assessment

The exact model and version in use, the retention and training position we hold with the model provider, the region inference runs in, and the transfer basis that applies, all change as contracts and model versions do. Rather than publish a snapshot that goes stale, we give them in writing, current at the date of the request, to any IG team assessing the product. Email info@thinktribal.com and ask for the AI processing schedule.

The same applies to the full sub-processor list, which is maintained alongside the data processing agreement rather than on a marketing page.

How the legal content is kept current

The rights and duties TenantSafe cites come from a curated set of current English legislation and statutory guidance: the Complaint Handling Code 2024, Awaab's Law and the regulations that implement it, the Housing Act 1985 and 1988, the Homes (Fitness for Human Habitation) Act 2018, and the consumer standards. The model is asked to apply that material, not to recall it.

When the law changes, the source material is updated and the change is dated. Awaab's Law Phase 2 on 30 November 2026 is the next scheduled change, extending the duty to seven further hazard types.

What we do not claim

• We do not claim the model never gets anything wrong. Language models can produce fluent text that is incorrect. The grounding above makes that much less likely and makes an error easier to spot, but it does not make it impossible.
• There is no human review step, by design. TenantSafe does not route a generated letter to a caseworker, a reviewer or anyone at ThinkTribal for approval before the resident can use it, and it is not going to. The resident reads what has been drafted and decides whether to send it, exactly as they would with anything they wrote themselves. If your assurance model assumes a professional signs off every outbound document, this product does not work that way and you should know that before you deploy it rather than after.
• We do not claim the product replaces legal advice. TenantSafe provides legal information. For advice on a particular situation, residents should contact Shelter or Citizens Advice, and take a solicitor's advice before any court action.
• An urgent hazard is not an AI matter. If someone is at immediate risk, the emergency route is the phone, not a complaint letter. TenantSafe surfaces the 24-hour emergency duty rather than drafting around it.

Asking us the hard questions

If you are assessing TenantSafe for an information-governance sign-off and something here does not answer your question, ask it directly. We would rather lose a deal at the assessment stage than win one on an assurance we cannot evidence.