Memorandum: from "trust us" to "check us"

Adopted by the founding members on 19 October 2026 as the shared philosophy of OVAIC. See dokument on avaldatud inglise ja veel viies keeles. Siduv on ingliskeelne tekst. English

1. Why now

AI systems now write, decide, pay and act on behalf of people. Two developments show where the world is going.

• The law. Since 2 August 2026, Article 50 of the EU AI Act requires providers of generative AI to mark synthetic content in a machine readable form so that it can be detected, with solutions that are effective, interoperable, robust and reliable as far as technically feasible. The law asks for proof that machines can check, not only labels that people read.
• The frontier developers. In September 2026 the chief executive of Anthropic, Dario Amodei, proposed that independent evaluators work inside AI companies to check "at the level of nuts and bolts whether an AI company is actually following the training, deployment, operational, and safeguards practices they claim to be following", with "the right to publish key findings about risk levels, incidents, practices, and the access they received or didn't receive", and to give "a second opinion free of commercial incentives". He also noted that a global body testing models before release is feasible, but that "the difficulty will be in verification".

Both point in the same direction: from promises to proof. Both leave the same question open: who checks, and how can anyone check the checker? Disclosure is necessary. It is not enough. A statement becomes trustworthy only when others can verify it without relying on the word of whoever made it.

2. Our philosophy

1. Check, do not trust. A claim about an AI system counts when it comes with evidence that others can verify: what data was used, who signed it, whether the sources are independent, and whether the result can be reproduced.
2. Interoperable or useless. Marks, signatures, provenance records and logs from different makers must work together. One vendor's proof that no one else can read is not proof.
3. Reproducible in public. Methods and test results are published with code and data. Anyone may rerun them.
4. Errors are a gift. Mistakes are corrected openly, with credit to the finder. History is not rewritten; corrections are added to it.
5. Independent eyes. Checks are done by parties free of commercial ties to what they check, and they may publish what they find.
6. No one is above the test. Members, including the initiator, run their own methods on the same public cases, and results against a member are published like any other.
7. We measure, we do not certify. OVAIC publishes methods and results. It does not declare anyone trustworthy.
8. For future generations. We build lasting public value, open to all, not profit for a few.

3. What members commit to

1. Bring at least one method or one test case to the open test bed in the first year.
2. Publish results on OVAIC cases with version, date and reproduction instructions.
3. Accept public correction of their own results.
4. Use the OVAIC name only as the Name and Results Policy allows.
5. Follow the Antitrust and Competition Policy and the Intellectual Property Policy.

4. First joint work

1. Article 50 test suite: public cases that check whether machine readable marks and provenance records survive editing, compression and transfer between tools, and whether they can be forged.
2. Evaluator evidence: an open format in which independent evaluators record what they checked, what access they had and what they found, signed and time stamped, so that their reports can themselves be verified.
3. Agent permissions: cases where an AI agent may only act when verified evidence is present.

5. Open to everyone

1. Any person, any background. OVAIC is open to every person and every organisation, whatever their age, faith, nationality, origin, skin colour or appearance. Persons under 18 join with the consent of a legal guardian. Under the statutes, organisations join as members and individuals as observers, without a vote. The statutes are binding. An individual who proves useful through a mathematical, scientific or cultural contribution may receive an equal vote only by decision of the General Assembly, in the way the statutes provide. A financial contribution does not count for this.
2. No one stays outside for lack of money. Where a fee cannot be paid, an idea, the wish to contribute or a request for support is enough to apply. The Board considers every such request, from any person or organisation, with full seriousness and answers in writing, with reasons.
3. We help those who help everyone. We know the difficulties of those who start with little, and we recognise them quickly.
4. Science, not metaphysics or marketing. We separate what can be checked from what is only believed or sold. Together we recognise pseudoscience, the wish for fame at any price and attempts to turn this common work into a private business.
5. Money decides nothing here. What lasts is verification, trust and good work for the generations after us. This is not only a philosophy. It is a choice between good and harm, and we make it openly.

Section 5 was added on 1 November 2026 to draft 0.1.

6. Vote

The founding meeting is asked to adopt this memorandum as the shared philosophy of OVAIC. It does not bind any member beyond the statutes and policies.

Sources

• EU AI Act, Article 50 (transparency obligations), applicable from 2 August 2026: https://artificialintelligenceact.eu/article/50/
• Dario Amodei, "We must pace the frontier" (September 2026): https://darioamodei.com/post/we-must-pace-the-frontier
• Anthropic, proposal for a frontier AI transparency framework (July 2025): https://www.anthropic.com/news/the-need-for-transparency-in-frontier-ai