Representation of the People Act & the Model Code
They bind candidates and parties. What we do: check every publishable line for the speech that gets cases filed, and block rather than release when the check cannot run.
Trust and safety
Political AI has earned its bad name honestly: bot networks, micro-targeting, synthetic video. We built the opposite on purpose, and this page is the part you can check.
Every story is bucketed deterministically before any advice is written. A counter cannot be assembled against an attack that was never aimed at you, and a judgement that would have you defending a rival is regenerated once and then dropped.
Every publishable line is checked for communal and caste appeals and for unsubstantiated allegations, in all four languages we write in. If the check cannot complete, the line does not go out.
Briefs carry source tags. Numbers carry a chip saying whether they come from the official record or from a model. A claim we cannot cite is dropped, not softened.
There is no code in ASMI that can post to a social platform. It drafts; your team publishes. This is a property of the build, not a setting.
Consent gates every message. One command erases you. Identifiers in our analytics are hashed, and nothing you send is ever used to serve another client.
We will not run intelligence for two rivals in the same race. It is a clause in your contract, not a line on a website.
Where we sit in the law
Most of these rules bind you, not us. We build so that using ASMI makes them easier to keep — and we do not claim approvals that do not exist.
They bind candidates and parties. What we do: check every publishable line for the speech that gets cases filed, and block rather than release when the check cannot run.
They direct political parties to label AI-generated and digitally enhanced campaign material. What we do: label the images we generate, at render, with the date — so your team can meet the advisory without remembering to.
Their labelling and provenance duties fall on platforms and intermediaries. We are neither. Where we generate an image it still carries a visible label and embedded provenance — the marks those rules expect a platform to look for.
We are a data fiduciary and we hold your data on consent. We are built to the Act, and building to the obligations that commence in 2027. There is no badge to hold up here, and we will not imply one.
The list we hold ourselves to
No voter databases and no micro-targeting. No messaging to voters. No bot networks, no fake accounts, no purchased amplification. No synthetic likeness or voice of any person, living or dead. No forecasts dressed up as findings. No communal or caste appeal, in any language. And no work for both sides of the same contest.
If a prospective client wants one of those, we are the wrong supplier, and we would rather say so at the first meeting.
We collect the phone number and display name of authorised subscribers, the messages they exchange with the agent, and what the agent produced. We do not collect voter rolls, identity documents, biometrics or precise location, and we never will. Retention, purpose and your rights are set out in full in the Privacy Notice, including how to erase everything in one command.
Transport encryption everywhere, signed webhooks, role-based access, an audit trail on administrative actions, and spending ceilings on every model provider so a fault cannot become an invoice. Kill switches exist on every scheduled surface and are tested.
Our Grievance Officer is named in the Privacy Notice, with the acknowledgement and resolution windows we hold ourselves to. If you have found a security issue, write to grievance@asmi.live with the detail and we will come back to you.