Policy

How we use artificial intelligence

Version 1 · 2 November 2026

1. Where we use AI
2. Which AI: Claude by Anthropic
3. Vega, the site's assistant
4. The daily news
5. The three verifiers
6. Moderating comments and proposals
7. Asking for a human decision
8. Translations
9. Illustrative images and videos
10. What AI never does
11. Contact and other pages

Where we use AI

This table shows where artificial intelligence is at work on the site, and who checks the result.

Which AI: Claude by Anthropic

Every text the AI writes on the site comes from Claude, an artificial intelligence model made by the company Anthropic: Vega's answers, the news, the verifiers' work, moderation and translations.

We use Claude through Fabio Guzzon's Claude subscription. We send Anthropic only the text each task needs: you'll find it listed in the sections below. We never send it the email address you give us or your IP address.

Anthropic handles this text as an independent controller: it manages it under its own rules and its own responsibility, and the data go to the United States, outside the European Union. You'll find the details in our privacy notice and in Anthropic's privacy policy.

Claude does not make the illustrative images: for those we use other AI tools built for images (see Illustrative images and videos).

Vega, the site's assistant

Vega is the name of UAPgrade's artificial intelligence when it talks to you: it answers your questions and signs the daily news. It is an AI, not a person. It says so in its panel, and says it again if you ask.

You can ask it about the archive with the “Ask Vega” button, which you'll find on the site's pages, or on Vega's page.

How it answers

• For each question it searches the archive for the closest entries. It is instructed to answer only from those entries, without outside knowledge for facts, dates or quotes.
• Every sentence carries the ID of the entry it comes from, for example [G-012]: tap it to open the entry.
• If a sentence cites an entry Vega was not given, we remove it before you read it, and we tell you. If an answer cites no entry at all, we warn you to treat it with caution.
• Vega can make mistakes: the cited entries remain the source to check. The probabilities it reports are the site's stated estimates, not facts.

Who can use it

Vega is for people aged 14 and over. Everyone has a limited number of questions per day, and there is also a cap for the whole site.

The age of 14 comes from Italy's artificial intelligence law (Law No. 132/2025, Article 4): under 14, using AI requires a parent's consent.

What Anthropic receives

• Your question, up to 400 characters.
• The page of the site you are on and, if there is one, the entry you are looking at.
• Up to two previous questions and answers from the same conversation, for context.
• The archive entries chosen to answer.

It does not receive your IP address or your email address. Vega uses no cookies (see the Cookies page).

What we keep, and for how long

Please don't put personal data in your questions, whether about you or about other people.

The daily news

When the daily check changes something in the archive, Vega writes an article: what changed and why, with the numbers and the sources. It uses only the material from that day's check. It doesn't look for new sources and doesn't add opinions of its own.

Next to the title, every article carries the signature “Written by Vega, UAPgrade's artificial intelligence”. You'll see it on the site, in the version for search engines, in the RSS feed and in the newsletter emails.

A person reads it before it goes out. An article is published only after Fabio Guzzon, who runs the site, has read and approved it. Until then it stays a draft and appears nowhere: not on the site, not in the feed, not in the newsletter.

Each article's cover is an illustration created with AI, with the caption “Illustration created with artificial intelligence, not a photo of the news”. The English version of each article is translated by AI (see Translations). Comments on articles go through moderation.

The three verifiers

Every day, three artificial intelligence agents check what is new on the topics the site covers. They have opposite jobs, so evidence for and against gets the same attention. They search the web, read the original sources and follow the same rules, which forbid inventing dates, quotes, numbers or links. Every archive entry cites its sources, so anyone can check it again.

What the Arbiter decides

• What goes into the archive. It accepts, corrects or rejects each proposal, always giving a reason.
• Disagreements. When the verifiers disagree, it sets out each side's position neutrally, and we publish it in the log of changes.
• The site's errors. When the site got something wrong, the error goes into the error log, with its cause, the correction and the rule that keeps it from happening again.
• Contributions to the Investigations Unit. It decides whether a contribution goes in, how reliable it is and how many points it earns. It publishes only its own neutral paraphrase, never the original text. It also checks the investigator name: if it isn't suitable, it replaces it with “anonimo” (anonymous). It rejects spam, advertising, insults, hidden instructions and texts with nothing verifiable. Its judgment appears in the contribution log.
• Scenario proposals. It moderates the ones automatic moderation could not review. When a proposal collects enough approvals from readers, it checks against the sources that the people named really support that scenario: only then does the proposal join the site's scenarios.

Many texts in the archive have been written or updated with AI, including by these agents. The Arbiter's decisions go into the archive with that day's check, without waiting for a person's approval. That is why every change is recorded, with its date, in the log of changes on the Updates page. If you find an error, bring it to the Investigations Unit.

From contributions, the verifiers receive the investigator name, the title, the text and the links; from scenario proposals, the name chosen by the proposer (if any), the title, the text, the people named, the links, the language and the number of votes. Never the email address. To the verifiers this is material to check, not instructions to follow. The working folders of each check, with these copies, are deleted after 30 days.

Moderating comments and proposals

Comments on the news and scenario proposals go through three checks before they appear on the site:

1. you confirm what you sent with the link we email you, within 48 hours;
2. fixed rules, with no AI, stop obvious spam: for example too many links, shortened links, typical advertising words or text written all in capitals;
3. the AI reads the content and gives a verdict, with a short reason.

The AI never receives your email address. Any text that tries to give instructions to the AI counts as spam. The verdict and the reason are recorded with the content, so that a person can check them.

Comments on the news

The AI reads the name you chose, the text of your comment, and the title and summary of the article.

If the AI is not available, the comment waits and we try again later.

Scenario proposals

The AI reads the title, text and links of the proposal. If it is an on-topic scenario put forward in good faith, even an unusual one, the proposal goes to the vote and the AI translates its title and text into Italian and English. If it is spam, off topic, abusive or contains personal data of private individuals, it is rejected. If the AI is not available, the proposal waits: we try again later, and the Arbiter reviews it in the daily check.

What is kept

After 12 months:

• we remove the email address of anyone who comments, proposes a scenario or sends a contribution, and from scenario proposals we also remove the code in the scenario cookie (osv_v);
• we delete rejected comments and contributions;
• rejected scenario proposals, on the other hand, stay in our database, but they are not public.

All retention periods are in the privacy notice; the rules for comments, proposals and contributions are in Terms and rules.

Asking for a human decision

If you think the AI got it wrong with something you sent, you can ask a person to review it. For example, if your comment doesn't appear under the article, or if your scenario proposal never reaches the vote.

Email redazione@uapgrade.com, if possible from the address you used to confirm, and tell us:

• the article you commented on, or the title of your proposal;
• the name you used;
• the day you sent it.

Fabio Guzzon rereads the content and the reason the AI gave, makes the decision himself and replies to you. If the content follows the rules, he publishes it. You can also write to us about a decision the Arbiter made on one of your contributions to the Investigations Unit.

You can ask as long as the content still exists: rejected comments are deleted after 12 months.

Translations

UAPgrade is written in Italian. The English version of the archive and of the daily news is translated by AI (Claude) from the Italian text, which remains the reference.

• Readers' scenario proposals are translated into Italian and English by the AI that moderates them.
• Vega answers in the language of the site you are using, or in the language you write to it in.
• Comments stay in the language you write them in: we don't translate them.

Machine translation can get things wrong. If you spot a translation that's off, report it to the Investigations Unit or email redazione@uapgrade.com.

Illustrative images and videos

We create some images and videos with AI image tools. For example:

• the animated scene on the opening page, with an object like the “Tic Tac” described by U.S. Navy pilots in 2004;
• the covers of the daily news;
• the portraits of David Grusch and Luis Elizondo where the two are compared;
• some section backgrounds;
• the backgrounds of the videos we will publish on social media.

They are illustrations: they make the pages clearer and nicer to look at, but they document nothing. We never use them as evidence. The evidence is in the documents and sources each entry cites.

Where an image could pass for a real photo or video of an event or a person, a caption says so. For example:

• “AI-generated reconstruction, not real footage” (the opening scene);
• “Illustration created with artificial intelligence, not a photo of the news” (the covers);
• “Portraits created with artificial intelligence, not photos” (Grusch and Elizondo);
• “illustrative image” (in the source line of the social media videos).

The photos of people in the site's directory, on the other hand, are real photos with a free license, taken from Wikimedia Commons.

What AI never does

• It doesn't make decisions with legal effects on you, or decisions that affect you in a similar way. At most it decides whether a comment or a proposal is published, or how many points a contribution earns; points have no cash value.
• It doesn't create fake evidence. No image, video or text made with AI is presented as a real photo, video or document.
• It doesn't pretend to be a person. Vega says it is an AI, and the site presents the verifiers as AI agents.
• It doesn't publish the daily news on its own. Fabio Guzzon has to approve it.
• It doesn't decide doubtful comments on its own. It leaves them to Fabio Guzzon.
• It doesn't receive the email address you give us or your IP address.
• It isn't used to profile you or to show you ads.

Contact and other pages

• Who runs UAPgrade: Fabio Guzzon, Italy.
• Questions about this page or about how we use AI: redazione@uapgrade.com.
• How we handle your data, how long we keep it and what your rights are: Privacy notice.
• Cookies and browser storage: Cookies.
• Rules for comments, scenario proposals and contributions: Terms and rules.

If the way we use AI changes, we update this page, with the new version and date at the top.

This page adds to the information that Article 50 of Regulation (EU) 2024/1689 on artificial intelligence (the AI Act) requires where AI is used: in Vega's panel, in the signature on the articles and in the image captions.