AI Transparency Notice

Last updated: 14 August 2026

DOGER is an AI system. This page says where it is used, what it will not be used for, and how the AI-generated material on this website is marked — the transparency information required by Article 50 of Regulation (EU) 2024/1689 (the EU AI Act).

1. Who is responsible

DOGER is built and operated by Layer2State, Lisbon, Portugal. Layer2State is the provider of the DOGER system under the AI Act, and a deployer of the third-party general-purpose models DOGER calls. Questions about anything on this page: hello@doger.ai.

2. Where AI is used

Two places, and only two:

• The product. DOGER answers questions in natural language over a knowledge graph built from public records — property listings, wildfire history, statistics, weather, air and noise data. An agent interprets the question, queries the graph, and returns a written answer with its sources. Access is by invitation, at app.doger.ai.
• Four background videos on this website. The moving backdrops behind the four use-case panels were generated with a video model. Section 5 lists them.

The rest of this site — the copy, the figures, the product screenshots and the recorded demonstrations — is human-written and human-recorded.

3. You are talking to an AI system

Every answer DOGER returns is generated by an AI system. No human reviews an answer before you see it. Answers are produced by large language models supplied by third parties and reached through a model-routing provider; the reasoning steps and the database queries behind an answer are logged and can be shown on request.

Your prompts and the data in your session are not used to train those models.

4. What DOGER is for, and what it is not for

DOGER is decision support. It assembles evidence about places, parcels, buildings, markets and environmental exposure so that a person can make a decision faster and with the sources in front of them. A person makes the decision.

The system is not built for, not sold for, and not to be used for any of the following:

• Scoring or rating individuals on their social behaviour or personal characteristics.
• Biometric identification, biometric categorisation, facial recognition, or inferring emotions.
• Predicting whether a named individual will commit an offence.
• Evaluating the creditworthiness or establishing the credit score of a natural person.
• Deciding a person's eligibility for a public benefit, a job, a school place, insurance cover, or emergency assistance.
• Any automated decision that produces a legal or similarly significant effect on a person without human involvement.

Where a bank, insurer or public body uses DOGER, what DOGER supplies is evidence about an asset and its location — cadastral facts, valuation comparables, wildfire and flood exposure, statistical context for the surrounding area. Scoring the counterparty, and the decision that follows, stay in the customer's own systems and remain the customer's responsibility.

5. AI-generated visuals on this site

Four short background loops on the home page were generated with Google's Veo 3.1 video model. They are decorative: abstract motion behind a use-case panel, played muted, hidden entirely on small screens. None of them depicts a real person, a real place, a real organisation or a real event, and none is presented as a recording of anything.

Each of the four is marked as artificially generated in three independent ways, so that the marking survives being downloaded, re-hosted or read by a machine:

• Content Credentials inside the file. Each file carries a cryptographically signed C2PA manifest, issued by the generating service, recording that it came from a generative model. Google is a signatory to the EU Code of Practice on Transparency of AI-Generated Content.
• Structured data on the page. A schema.org block in the page source declares each file with digitalSourceType: TrainedAlgorithmicMediaDigitalSource — the IPTC term for fully machine-generated media.
• Attributes on the element. Each <video> tag carries data-ai-generated="true".

The disclosure is kept to this page and to the file metadata rather than stamped across the visuals themselves. Article 50(4) permits that for material forming part of an evidently creative work, where disclosure has to be made in a manner that does not hamper the display of the work.

By contrast, these are not AI-generated and carry no such marking: the product screenshots, the wildfire-forecast map loop, and the recorded product tour. Those are real captures of the running system.

6. What the system can get wrong

DOGER reports what is in the public record, and the public record is uneven. Sources are refreshed on different cycles, so figures can lag reality. Coverage varies between municipalities. Forecasts — wildfire probability in particular — express a likelihood, and are published together with the accuracy the model actually measured. Language models can also state something with more confidence than the underlying data supports.

So: every answer cites its sources, and every figure that matters should be checked against the cited source before anyone acts on it. Nothing on this site or in the product is legal, financial, insurance or investment advice.

7. Human oversight

DOGER is designed to be overseen: answers carry their sources and the queries behind them, sessions are logged and auditable, and a customer deployment can be run inside the customer's own environment. Users can challenge an answer, ask for the underlying records, or ask for a query to be re-run. Where a customer's own process attaches consequences to a DOGER output, that customer is responsible for the human review in that process.

8. Data

The graph is built from published public-sector and public-market sources. It is not built to profile individuals, and personal data is not retained in it unless a customer explicitly scopes it and a lawful basis exists for doing so. How this website handles your own personal data is covered in the Privacy Policy.

9. AI literacy

Everyone at Layer2State who builds, operates or demonstrates DOGER is briefed on what the system can and cannot do, how its outputs should be described to customers, and the obligations in this notice. New capabilities are reviewed against these boundaries before release.

10. Raising a concern

Write to hello@doger.ai and we will respond. The national authority supervising the AI Act in Portugal is ANACOM, which acts as the market surveillance authority and single point of contact — anacom.pt. For personal-data matters the supervisory authority is the CNPD — cnpd.pt.