Protocols
How AI:OK suspends and withdraws certification, how it handles confidential evidence and data, and how it secures its systems.
Last updated 12 October 2026.
Part 1. Revocation Protocol
How AI:OK suspends and withdraws certification.
1. Purpose
A trustmark has value only if it can be withdrawn. This protocol sets out when and how AI:OK suspends or revokes certification, so that certified entities, rights holders and the public know what to expect.
2. What this protocol decides
AI:OK decides whether a certified entity still meets the AI:OK standard on the evidence available. AI:OK does not decide whether any person has infringed copyright or any other right. That question belongs to courts and competent authorities. Where this protocol refers to a rights position, it refers to what the entity has shown AI:OK.
3. Grounds
Material misrepresentation. Evidence submitted at certification, renewal or in response to a request was knowingly or recklessly incorrect, or material information was withheld.
Rights position not substantiated. The entity cannot show, when asked, that it holds the rights it represented at certification. This includes continuing to use material after a valid opt-out, and using restricted, non-commercial or research-only data commercially without appropriate rights.
Governance not operating. Governance described at certification is not operating in practice, for example record keeping, disclosure on request or remediation steps that were described but are not carried out.
Material change not disclosed. A material change to training data, model structure, rights arrangements, ownership or control, third-party model dependencies or deployment markets was not disclosed promptly, or a change that required re-certification was not notified.
Remediation not completed. Conditional Certification was granted and the agreed remediation plan was not completed in the agreed period.
Trustmark misuse. The trustmark is displayed outside the certified scope, after expiry, suspension or revocation, in modified form, or in a way that suggests AI:OK endorses claims beyond the certified basis.
A seventh circumstance ends certification without a finding against the entity. If the annual review is not completed, certification lapses and the registry shows Expired, not Revoked.
4. Statuses
Certified. All mandatory requirements are met and current. The trustmark may be used within the certified scope.
Conditional. Certified with a remediation plan and a re-review date. The trustmark may be used only as the certificate states.
Under Review. A question has been raised and is being assessed. The trustmark may be used pending the outcome.
Suspended. A serious question is pending, or an interim measure applies. Trustmark use stops immediately.
Revoked. Certification is withdrawn. Trustmark use is prohibited.
Expired. The annual review was not completed. Trustmark use is prohibited.
5. Process
Trigger. AI:OK learns of a possible ground through an evidence filing, a regulator referral, its own monitoring or an annual review.
Acknowledgement. Receipt is acknowledged within five working days. Regulator referrals are handled first.
Preliminary review. An assessor decides whether the matter has substance. If not, the matter is closed and the submitter is told.
Notice. If the matter proceeds, the entity receives written notice of the specific ground and the substance of the evidence. Where a submitter asked to stay anonymous, AI:OK describes the allegation without identifying the submitter, unless fairness or the law requires otherwise. AI:OK tells the submitter before that happens.
Response. The entity has 20 working days to respond. AI:OK may extend this.
Determination. An assessor decides, or a panel where the matter is complex or sets a precedent. The decision applies the standard of proof in section 6. Written reasons go to the entity.
Publication. The registry entry is updated and a reason category is published (section 8). Underlying evidence is not published.
6. Standard of proof
AI:OK acts where, on the evidence, it is more likely than not that a ground applies. An allegation alone is not sufficient. A filing is treated as information to be tested, and the entity is given the chance to respond before any final decision.
7. Interim suspension
If AI:OK reasonably believes that continued use of the trustmark would mislead the market or cause harm, it may suspend before the process ends. The entity is told promptly, with reasons, and the process continues. The entity may respond to the suspension within five working days, and AI:OK reviews it within a further five working days.
8. Publication
Status changes are published on the registry. For suspension, revocation and expiry AI:OK publishes one of these reason categories: not renewed, misrepresentation, rights position not substantiated, governance not operating, material change not disclosed, remediation not completed, trustmark misuse. Evidence, submissions and investigation detail are not published. AI:OK does not publish a statement that a person has infringed any right.
9. Appeal
An entity may appeal a suspension or revocation within 20 working days of notice, by writing to legal@ai-ok.org. A person or panel not involved in the original decision considers the appeal. The appeal outcome is added to the registry entry. A suspension stays in place during an appeal unless the appeal body decides otherwise.
10. Reinstatement and reapplication
A suspension may be lifted when the underlying matter is resolved. A revoked entity may reapply. Reapplication is a full new assessment, not a reversal, and the registry keeps the history.
11. Who can file
Rights holders and their representatives can file complaints. Regulators and competent authorities can make referrals, which AI:OK handles first. Current and former staff of a certified entity, and other people with direct knowledge, can make a confidential disclosure through a separate secure channel.
AI:OK handles confidential disclosures confidentially. AI:OK is not the discloser’s employer and is not a competent authority, so it cannot provide the legal protections that employment law and whistleblowing law give. A person who is unsure of their position should take independent advice before filing.
12. Filing in bad faith
AI:OK may decline to act on filings that are knowingly false, abusive or made to gain a commercial advantage over a certified entity, and may restrict further filings from the same source. Filers confirm on the form that the information is accurate to the best of their knowledge.
13. Confidentiality
Filings and material considered under this protocol are handled under the Confidentiality Protocol.
14. Changes to this protocol
Changes take effect 30 days after publication, unless a change in law requires earlier effect. Matters open at the date of change continue under the version in force when they began.
Part 2. Confidentiality Protocol
How AI:OK handles the evidence and commercially sensitive material submitted during certification, including how it is used, who sees it and what is published.
1. Purpose
Certification requires applicants to disclose commercially sensitive information: training data sources, licence terms, supplier relationships, model architecture and rights positions. This protocol says how that material is handled. The standard works only if applicants can submit real evidence without taking on commercial risk. The Privacy Policy covers personal data. This protocol covers the evidence itself.
2. What it covers
All material submitted to AI:OK for a Quick Check, waiting list entry, Core submission, renewal, remediation, Extended Framework review or evidence filing. This includes questionnaire answers, the Dataset and Source Register, licence agreements, supplier statements, legal opinions, model and provider details including versions, remediation records and supporting documents. For catalogue certification it also includes catalogue manifests and per-work rights status.
3. Marking material confidential
You may mark any answer or attachment [Confidential]. Unmarked material is also treated as confidential business information. Marking tells assessors that you consider the material particularly sensitive.
4. Who has access
The assessor assigned to the matter, and panel members where it is escalated.
A small number of AI:OK technical administrators who maintain the systems that store the material. They access it only where needed to operate, secure or repair those systems, and they are bound by written confidentiality obligations.
Nobody else has access. That includes other applicants, certified entities, commercial partners, channel partners, advisers, funders, institutional partners and AI:OK staff outside the review process. An assessor with a material relationship to the applicant is recused and does not receive access. Reports to funders and institutional partners use aggregated figures, such as the number of applications and outcomes. They do not include applicant evidence.
5. How evidence is used
Evidence is used only to assess the application, maintain the certification and, where relevant, consider a revocation matter. It is not used for any other purpose without your separate written consent.
AI:OK does not use submitted material to train any AI model, its own or anyone else’s. AI:OK does not sell, license or share submitted material with competitors, other applicants or commercial partners.
The legal basis for processing any personal data in it is contract. Where the material relates to a revocation matter, the basis is legitimate interest in the integrity of the scheme.
6. What is published
If certification is granted, the registry shows only the entity name, certificate number, status, scope, issue date and review date.
If certification is suspended, revoked or expires, the registry also shows a reason category from the Revocation Protocol. Evidence is not published.
No dataset, licence term, supplier name or commercial arrangement is published without your written consent.
7. Rights holders and other third parties
If a rights holder or other third party contacts AI:OK about an entity, AI:OK does not confirm or share what that entity has submitted. AI:OK may tell the entity that a filing has been made and may describe its substance as the Revocation Protocol allows.
8. When AI:OK may disclose
AI:OK may disclose confidential material only:
where the law or valid legal process requires it;
where a regulator with jurisdiction over AI:OK requires it;
where it is reasonably needed in a revocation process, limited to the minimum necessary and to the parties involved; or
where you have consented in writing.
Where the law allows, AI:OK tells you before disclosing.
9. Where it is handled
AI:OK is based in Ireland and many applicants are elsewhere, including the United Kingdom and the United States. Submitting evidence will usually move data from your country into the European Economic Area, and some of our service providers are in the United States. We apply one standard to all evidence, wherever the applicant is based. The Privacy Policy lists the providers and the transfer safeguards.
10. Security and incidents
Part 3 of this page describes how material is stored and protected. If AI:OK confirms that confidential material has been accessed or disclosed without authority, it tells the affected applicant or certified entity without undue delay, whether or not personal data is involved.
11. Integrity
AI:OK records a cryptographic hash of each evidence file on receipt, so both parties can confirm exactly what was submitted if a question arises later.
12. Retention
Material is kept for the duration of certification and for six years afterwards, to meet record-keeping needs and to support any later audit, dispute or regulatory enquiry. Material from applications that do not proceed is kept for two years in case of resubmission or appeal, then deleted or anonymised.
13. Your rights
You may ask for a copy of what you submitted, correct inaccurate information before a decision is made, and ask for deletion once the certification relationship has ended, subject to the retention periods above.
14. Contact
Part 3. Security
AI:OK runs its certification process on established third-party platforms rather than custom-built infrastructure. This page describes the security measures in place today and the responsibilities that sit with you.
1. Encryption
Data sent through our website, questionnaires and evidence forms is encrypted in transit using TLS. Data at rest is protected to the standard of each platform provider. Each provider publishes its own security and compliance documentation.
2. Where data is held
Application and evidence records are held in Airtable.
The website, member areas and registry are hosted by Squarespace.
Payments are processed by Stripe, which is PCI DSS compliant. AI:OK does not store full card numbers.
Support chat is provided by Intercom or Tidio.
The Privacy Policy lists any other provider that processes personal data.
3. Access controls
Access to application and evidence records is limited to assigned AI:OK assessors and technical administrators, who sign in through each platform’s own authentication. We apply least privilege: an assessor sees the matters assigned to them rather than the full set of records by default. An assessor with a material relationship to an applicant is recused and their access is removed. Access is reviewed when people join, change role or leave.
4. Evidence integrity
Every evidence file is hashed on upload. This gives you and AI:OK a verifiable record of exactly what was submitted and when.
5. Suppliers
We select suppliers that publish security documentation and sign data processing agreements.
6. Incident response
If we become aware of a security incident affecting personal data, we notify the Irish Data Protection Commission within 72 hours where the GDPR requires it, and we notify affected people without undue delay where the incident is likely to result in a high risk to their rights and freedoms. If confidential evidence is accessed or disclosed without authority, we tell the affected applicant or certified entity without undue delay, whether or not personal data is involved.
7. Reporting a vulnerability
If you think you have found a vulnerability affecting AI:OK, email security@ai-ok.org with enough detail to reproduce it. Please do not access, change or delete data that belongs to others, do not disrupt the service, and give us a reasonable time to fix the issue before you publish anything. We will acknowledge your report within five working days.
8. Your responsibilities
If you have portal access, keep your credentials confidential and use them only for your own organisation. You are responsible for activity under your account. Tell us at security@ai-ok.org if you suspect unauthorised access.
9. Contact