Five questions about the watch.
01- What does Vigilia actually do?
- It builds instruments that observe AI as it is deployed, keeps public records of what they find, and publishes them in five languages. The register visits AI systems and records whether they say they are AI; the Sentinel records every action an AI agent takes and seals the record; the warnings register keeps what those who build AI have said about where it is going.
02- Who can be recorded?
- Organisations, and the public professional roles that speak for them — never a private individual. Vigilia visits the public page of organisations that run a chat or assistant surface. If a person joins the conversation the visit stops before their words are captured, and any personal detail caught incidentally is redacted before anything is published. Vigilia's own site is record 1 and is held to every rule that applies to the others.
03- How do we reply to a record, or correct it?
- Write from an address at your own domain to observer@aivigilia.com. A reply is published beside the record verbatim, in your language, dated and unedited. A correction is made on evidence — a wrong organisation, a mis-identified vendor, a mis-read reply, a fault in the visit — and appears in the corrections log, dated, with the original left visible. Claiming a record is free; it lets you point the observer at the page where your surface lives and ask for one re-check after a fix.
04- How do we decline a visit?
- A disallow for VigiliaObserver in your robots.txt ends a visit before it starts, and it is never bypassed. A written request from an address at your organisation to observer@aivigilia.com is honoured in exactly the same way. The record then shows one line — organisation declined automated observation, with the date — which is a factual state, not an adverse finding.
05- What is the free self-check, and what is it not?
- It is a tool with no price. In one mode it queues one identified visit to a public website under the observer's rules; in the other it maps a system you describe to the EU AI Act — the risk tier, the articles that govern it, and the gaps it can see — using the open rulebook and a published, deterministic scoring. It reads what you describe, not the system as it runs. It is a gap analysis, not legal advice and not a certification, and nothing it produces is a statement about anyone's legal position.