Skip to content

We've published our Article 50 transparency declaration, with the table of what still isn't done

We've published our Article 50 transparency declaration, with the table of what still isn't done

There's a sentence that opens our transparency declaration and explains why it exists: Article 50 was already being met by design in what we had published, but it wasn't written down anywhere, and what isn't written down can't be shown. That was the gap, and it's the one this document closes.

It's worth saying straight away what it is and what it isn't. It's a self-declaration, signed under our own responsibility, that can be shown for as long as it's true. Nobody audits it and nobody issues it. It isn't a certificate, not least because for a conversational assistant none exists. It's published in Spanish, at aiginer.com/declaracion-transparencia.

The four obligations, and whose each one is

Article 50 splits obligations between whoever develops an AI system and whoever deploys it, and they aren't the same. We're on both sides: we deploy our own systems on our website and we build systems our clients deploy. The declaration lays them out in a table, paragraph by paragraph, and says which specific thing of ours each one lands on: the website assistant, generated images and video, avatars, and published text.

What we do, and it can be checked

Anyone writing to the assistant on aiginer.com knows from the first message that they're talking to an automated system, and when the assistant can't resolve something it says so and offers a person. There is no point in the journey where somebody gets stuck with a machine.

It doesn't keep conversations either, and that's a design decision rather than an oversight: there's no history to consult, ours or anyone's. The conversation doesn't leave the European Union, and model providers are expressly instructed not to use that data for training.

With dates, not adjectives

The declaration includes a table of next steps, each with its status and its deadline: machine-readable marking of generated content, checking it in our agents widget, and bringing the written breach-notification procedure into the processing agreement. When one is closed, the table is updated and the date is added.

A declaration that only lists adjectives isn't transparency, it's advertising. That's why the table sits inside the document, in plain view of whoever reads it.

What we are not, written once

  • We aren't certified in anything. Not ISO 27001, not ISO 42001, not SOC 2. We'd rather say it than hint at a process under way.
  • This isn't an AI Act certification. None exists for limited-risk systems like a conversational assistant. What applies is the transparency obligation, and that's what's declared.
  • This isn't legal advice. It's what AIGiner does with its own systems.

There's also a point almost nobody gets right and the declaration states: in Spain there is still no penalty regime developing these obligations. Meeting them is mandatory all the same, but anyone selling this by waving the size of a fine around is overstating it.

And a second thing the same day: certificates can be checked

Whoever finishes our course gets a certificate with a serial number, and from today that number can be checked. You type it into the academy's verification page and it says whether it's on record. It's public and it asks for no email and leaves no trace: a certificate only the issuer can confirm isn't much use when the person asking is your own client.

What we ask of whoever deploys a system of ours

When we build an agent for a client, the client is the deployer and part of Article 50 becomes theirs. With every delivery we hand over, in writing, what notice they have to show and where, what gets logged and what doesn't, who the responsible human they appoint is, and the list of providers involved. That isn't an extra: delivering a system without it would be handing someone a problem wrapped in a product.

The principles behind all of this are in our AI policy.