Legal
How we use AI, and how we say so
Three rules, applied as gates in the publishing path rather than as good intentions.
The three rules
Real
Nothing invented is presented as real. No fabricated testimonials, statistics, people or places. If a number appears on this site or on a client’s, it came from a record and the record names its source. If we have not got a number, we say so rather than reaching for one.
Disclosed
Where a model was involved in something you read or watch, and the EU AI Act, the Advertising Standards Authority or the platform requires it to be said, we say it, on the thing itself. Every journal entry carries a line at its foot describing how it was written. Images and video we make carry a C2PA manifest naming the tool and the person responsible.
Audited
No copy is published until the thirteen-check audit has passed and a certificate exists. The checks are: truthful claims with a source; nothing invented; AI involvement disclosed where required; no unsubstantiated superlatives; prices carrying the VAT position; no dark-pattern language; plain, accessible language; British English and our voice; correct legal names and company details; rights to every asset named; no personal data without a basis; links that resolve and say where they go; and the client’s own facts matching their record.
What we use models for
- Drafting: first drafts of copy, social posts and journal entries, always edited by a person.
- Classifying: sorting a brief by the kind of business and the area it is in.
- Summarising: turning a month of Ledger entries into a report a person then checks.
- Craft: image work, subtitles and edits around real footage and real photographs.
- Routine work inside our own platform: triage, drafting replies, watching for changes.
What we do not use models for
- Inventing a quote, a review, a statistic, a person or a place.
- Making a decision that needs a person’s yes. A model never decides it has been approved.
- Publishing anything without a person reading it first.
- Generating a photograph of premises, staff or work that we present as a photograph.
- Writing this page, the terms, the privacy notice or the accessibility statement.
The ladder
We build at the lowest rung that does the job: code first, then algorithms, then small models, and large models last. A change that adds a large-model call has to answer, in writing, what the model does that code cannot, what it costs per unit, what its cap is, how its output is validated and what happens when it fails. If it cannot answer, it is not built that way.
Budgets and safety
Every agent has a profile, a set of tools and an approval policy, and every unit of work has a hard cap in pounds: 3 per cent of a project’s price, 10 per cent of a Care fee, 8 per cent of a Search or Social fee, and 15 per cent of a module’s price. A run that would exceed its cap stops, writes the entry and asks a person, in pounds.
Model output is untrusted input. Instructions and data are kept apart at the boundary, anything a model returns is validated against a schema before it touches a database, a file or another tool, and no tool receives text from a model that a person or a validator has not bounded. Agents write an entry before and after every tool call.
The words on the home page, the services pages, the pricing page, the legal pages and this one were written by a person. Journal entries are drafted from the Ledger with a model’s help, edited by Joe Kaul, and carry their disclosure line at the foot. Case studies are written from the client’s Ledger and approved in writing by the client before they are published.
Telling us we got it wrong
If you think something on this site is not what it says it is, write to hello@mail.kalotect.com. We will check it against the record, correct it if it is wrong, and say on the page that we did.