AI Disclosure and
Editorial Standards

We are an AI-native company. This is exactly how we use AI, what we do not do with it, and who is accountable.

1. What "AI-native" means here, and what it does not

We call ourselves AI-native because AI is in the default path of how we work, not bolted onto the side of it. That is a statement about our method, and it is verifiable.

It is not a claim about capability. Three things "AI-native" is sometimes used to imply would be untrue of us:

• We do not develop foundation models. We build on general-purpose models supplied by third parties — currently Anthropic (Claude), OpenAI, Google (Gemini) and Mistral AI. Where we describe something as ours we mean the application, the workflow, the prompt and agent architecture, the evaluation harness and the judgement around it — not the model.
• We do not claim autonomy we do not have. Where we describe a system as agentic we mean it executes multi-step tasks under human supervision with defined limits, not that it operates unsupervised.
• We do not attach numbers to AI performance unless we can produce the measurement. Any quantified claim we publish about speed, accuracy, cost or outcome has a dated substantiation record behind it, and we will show it to anyone who asks.

We hold ourselves to this because the honest version is more useful to you, and because overstating AI capability is a recognised category of misleading advertising. Regulators have enforced against it: the US Securities and Exchange Commission fined two advisers a total of USD 400,000 in March 2024 for misrepresenting their AI use, and the US Federal Trade Commission has pursued a series of exaggerated AI claims since September 2024 — including a case where the offence was a single unsupportable accuracy figure. In Switzerland, incorrect or misleading statements about one's own business or services are actionable under Article 3(1)(b) of the Unfair Competition Act, and under Article 13a a court may require the advertiser to prove its claims.

2. Where we use AI

3. Where we do not use AI

• We do not publish AI output that no human has read and checked.
• We do not generate testimonials, client quotations, case studies, reviews or endorsements. Everything attributed to a named person or organisation is real, said by them, and used with their permission.
• We do not generate photorealistic depictions of real people, organisations, places or events, and we do not synthesise anyone's voice or likeness. See section 6.
• We do not fabricate data, sources, citations, statistics or research findings, and we do not present synthetic or simulated data as evidence.
• We do not let an AI system make a decision about a person — a candidate, a client, a founder, an investor — without a human deciding.
• We do not enter confidential client or personal information into tools that are not approved. See section 8.

4. Human review and editorial responsibility

This is the load-bearing commitment on this page.

Substantive review means a deliberate examination of the substance by a person with relevant knowledge and professional judgement, with authority to approve, alter or reject the text on substantive grounds including fact-checking. It is not a spell-check and it is not a formality. Before anything is published:

1. a named human being has read it in full;
2. every factual claim, figure, statistic, quotation and citation has been verified against its primary source;
3. any claim we cannot substantiate has been removed or qualified;
4. a named person has approved publication and is accountable for it.

We keep a record for each substantive publication: the human author, the AI tools used and for what, the reviewer, the date, and what was checked or changed. That record is our evidence, and it also supports our position on ownership in section 12.

Editorial responsibility is held by Fabien Lopez, Founder, reachable at [email protected].

This is also the basis on which we rely, where relevant, on the exemption in Article 50(4) of the EU AI Act for AI-generated text that has undergone human review and editorial control and for which a person holds editorial responsibility. That exemption reaches text only. It does not extend to synthetic image, audio or video, which we disclose under section 6 regardless of how thoroughly they were reviewed.

5. How we label AI-assisted content

We use the phrase "AI-assisted, human-reviewed" rather than "AI-generated", because it is the accurate description of our process. Research on how audiences read these labels finds that "generated by" is understood to mean entirely machine-written, while "assisted" is understood as human–machine collaboration — and that a visible commitment to human review is the strongest single driver of credibility. We are not choosing the softer word; we are choosing the true one, and we accept the burden in section 4 that makes it true.

Our practice:

• Substantive publications carry a production note stating that the work is AI-assisted and human-reviewed, and naming the accountable person.
• Labels appear at the top of the content, not buried at the end.
• Where content is substantially machine-drafted with limited human authorship, we say "AI-generated" and not "AI-assisted".
• Ordinary navigation and product copy is not individually labelled; this page states the general position for the whole site, and our footer carries the same statement.

6. Synthetic media, images and voices

We commit, as a matter of policy and not merely of compliance:

• We do not create AI-generated or AI-manipulated images, audio or video that depict real people, organisations, places or events in a way that could be taken for authentic. No synthetic team photographs, no synthetic client footage, no cloned voices, no fabricated events.
• Where we use AI-generated illustrative or abstract imagery — diagrams, patterns, conceptual artwork that depicts nobody and nothing real — we do not consider it a deep fake, and we say so here rather than labelling each image.
• If we ever published synthetic media resembling real people or events, we would disclose it clearly and immediately adjacent to the content, at first exposure.

Declared synthetic media

Two published assets sit against the first commitment above, so we name them here rather than let the general policy imply they do not exist. They are not equally attested, and the difference is stated rather than smoothed over.

• BuddyLeader market film — a ten-second consumer film on our Ventures page. It is AI-generated, not AI-assisted: the file carries a C2PA content-credentials manifest signed by Google, recording “Created by Google Generative AI”, a trainedAlgorithmicMedia source type and an imperceptible SynthID watermark. We serve the original file unaltered so that manifest survives and anyone can verify it. The people in it are synthetic and depict no real person; the tradespeople shown are not customers and the film is not client footage. The application interface shown on screen is simulated and is not a recording of the live BuddyLeader product. The film is labelled AI-generated · simulated interface · dramatisation immediately adjacent to the player, at first exposure, per the commitment above.
• SELF-MADE, the launch film — a thirty-one-second film on our home page. It is AI-generated, not AI-assisted. Unlike the BuddyLeader film it carries no C2PA manifest and no cryptographic provenance of any kind: we examined the file and there is no content-credentials record in it. This declaration is therefore our word, not a signature you can verify, and we would rather say that than let the paragraph above imply otherwise. The film depicts no person, real or synthetic, and no product interface: it is a stone, a blossom, a root system and a Swiss landscape. Nothing in it is offered as a record of a real place, event or result. It is labelled AI-generated · dramatisation · no cryptographic provenance manifest immediately adjacent to the player, at first exposure. The film carries a score and no speech; every word it contains is an on-screen title, and all of them are printed in full beneath the player so that a visitor using a screen reader receives the same content as a visitor watching it.

We are replacing the simulated interface with a recording of the real product. Until that is done, the label stands and the film is not offered as evidence of what the product does — the verified claims about BuddyLeader live on Facts.

This aligns with Article 50(4) of the EU AI Act on deep fakes, which has no public-interest limitation and applies to commercial content, and with the Swiss FDPIC's position that programs enabling the manipulation of faces, images or voices must always be clearly indicated.

7. Accuracy, hallucinations and corrections

AI systems produce fluent, plausible text that is sometimes simply wrong. Confident invention of sources, statistics and quotations is a known failure mode, and fluency is not a signal of accuracy. We do not treat this as an acceptable cost of using AI; we treat it as the specific thing our review process exists to catch.

• Every figure, statistic, date, quotation and citation is checked against its primary source before publication. Aggregators are not primary sources.
• Every published figure carries its vintage, its denominator and its source, and third-party benchmarks are labelled as benchmarks and never presented as our own results.
• Every quantified claim about our own performance has a dated substantiation record on file before it goes live.
• Facts about named third parties are independently verified by a human before publication.
• Where a figure fails our standard we name it in public rather than quietly omitting it.
• Corrections are public. Tell us something is wrong at [email protected]. We aim to assess a challenged figure within 24 hours and to record substantive corrections on the page that carried the error.

None of this is a disclaimer. If we publish something wrong it is our error and our responsibility, not the model's. We do not offer "the AI did it" as an explanation, and we do not think it is one.

8. Confidentiality and client data

Feeding material into a third-party AI tool is a disclosure to a third party. We treat it that way.

• We use an approved list of tools on business or enterprise terms under which the provider is contractually bound not to train on our inputs. We verify this in writing per tool and re-verify it when terms change.
• A written data processing agreement is in place with each AI provider that handles personal data on our behalf.
• We do not enter client confidential information, personal data or commercially sensitive material into tools outside that list. Where we work on client material we minimise or pseudonymise identifying detail.
• We check our confidentiality obligations before any client material goes into any tool. Where an agreement restricts disclosure to third parties or requires consent for sub-processors, we obtain consent or we do not use the tool.
• Clients may restrict our use of AI on their engagement, including prohibiting it entirely. Ask, and we will record it in the engagement terms and honour it.
• Processing locations are the European Union, Switzerland and the United States; cross-border transfers are handled as described in the Privacy Notice.

We do not claim that no data ever leaves Switzerland, and we do not claim that we never use AI on client work. Both would be untrue, and a false confidentiality claim would be worse than the thing it conceals.

9. Data protection and your rights

Swiss data protection law is technology-neutral and applies directly to AI-supported processing; the FDPIC has said so repeatedly, most recently in its update of 8 May 2025. Accordingly:

• You have the right to know whether you are corresponding with a machine or a person, and we will tell you.
• We do not use your personal data to train AI models, and we do not permit our providers to.
• We do not take decisions about you based solely on automated processing. If that changes you will have the right to state your position and to have the decision reviewed by a human being — FADP Art. 21, GDPR Art. 22.
• You may object to processing and exercise your other rights as described in the Privacy Notice.
• Where we assess that an AI use case carries high risk to individuals, we carry out a data protection impact assessment before deploying it.

10. AI in our products

Where we build AI features into ventures we accept that we may be the provider of those AI systems and not merely a user of someone else's, with the heavier obligations that follow. In those products:

• users are told clearly, before or at the first interaction, that they are interacting with an AI system;
• the system's purpose and its limitations are stated plainly;
• the system does not present its output as professional advice — medical, legal, financial or psychological — and says so;
• there is a route to a human being, and a way to report a problem;
• we monitor output quality and we are accountable for what the system tells users. A tribunal in Moffatt v Air Canada (2024 BCCRT 149) held a company liable for its chatbot's incorrect statements and rejected the argument that a chatbot is a separate entity from the business operating it. We think that is obviously right, and we do not intend to argue otherwise.

11. Training and AI literacy

Everyone working for Living Scale Up — employees and contractors — receives role-appropriate instruction on the AI tools we use, what they are good at, how they fail, and the confidentiality and verification rules in this document. This is governed by a written internal AI policy covering approved and prohibited tools and uses, confidentiality, the verification duty, and escalation. Training is refreshed at least annually and when the policy changes, and we keep a dated record of it.

This is also how we meet the AI literacy obligation in Article 4 of the EU AI Act, in force since 2 February 2025.

12. What AI use means for ownership

An AI-native company should be honest about this rather than quietly assert the maximum.

Copyright requires a human author. In Switzerland the Federal Institute of Intellectual Property states plainly that copyright protection applies only to works created by humans, and draws the line between AI as a tool and AI as the creator. Courts in Germany and elsewhere have reached the same conclusion, holding that however elaborate the prompt, purely AI-generated output is not protected. The US Copyright Office has held that prompts alone are not authorship.

• We assert copyright in the human-authored elements of our publications — the conception, structure, argument, analysis, judgement and expression our people contribute.
• We do not assert that everything on this site is a protected copyright work, because for material that is substantially machine-generated that would not be true.
• We document human contribution for the work that matters, both because it is the condition of protection and because it is the evidence for the review process in section 4.
• Our trade marks do not depend on authorship and are unaffected by how a name or logo was produced.
• Our frameworks and methods are not protected by copyright in any event — copyright protects expression, never ideas or methods. Anyone may learn our methods and practise them. What we ask is that they not use our names for their version of them.

The full position is in the Intellectual Property, Trade Marks and Citation Policy.

13. Standards we align to

• EU AI Act, Regulation (EU) 2024/1689 — Article 4 on AI literacy, in force since 2 February 2025, and the Article 50 transparency obligations, applicable from 2 August 2026. The Digital Omnibus amendment, Regulation (EU) 2026/1744, deferred certain high-risk obligations; it did not defer Article 50. We assume the Act applies to us: although we are established in Switzerland, Article 2(1)(c) reaches third-country deployers where the output of the AI system is used in the Union, and Article 2(1)(a) reaches us as a provider for any product we make available in the Union.
• European Commission guidelines on Article 50 transparency, adopted 20 July 2026, and the Code of Practice on Transparency of AI-generated Content, published 10 June 2026 and assessed adequate in July 2026.
• Swiss FADP and the FDPIC's published positions on AI and data protection.
• Council of Europe Framework Convention on Artificial Intelligence, which Switzerland signed on 27 March 2025. It binds states rather than companies, but its principles inform how we work.
• Swiss Unfair Competition Act Article 3(1)(b) and Article 13a — the discipline that every claim we make must be provable.

Switzerland has no horizontal AI statute. The Federal Council decided on 12 February 2025 on a sectoral approach and to ratify the Council of Europe Convention, with a consultation draft expected by the end of 2026. Until then, and probably for some years after, the EU AI Act is the binding external constraint on us, and we plan on that basis.

14. Who is accountable

This policy is reviewed at least every six months and whenever the law, our tools or our practice change materially. The version and date are at the top of this page.

If we breach it: a breach is treated as a serious matter, investigated by the governance owner above, and remediated — including public correction where published content is affected, notification where a confidentiality or data protection obligation has been broken, and disciplinary or contractual consequences where appropriate. We say this because most published AI policies do not, and a commitment with no consequence attached is not a commitment.

15. What we do not claim

We do not claim this is perfect, finished, or better than anyone else's. Our practices will change as the tools, the law and our understanding change. We will get things wrong, and when we do we would rather find out quickly.

If you think something here is wrong, incomplete, or contradicted by something we have actually published, write to [email protected]. We will answer.