Trust center
Trust & provenance
Evidence over adjectives. This page describes the methodology behind every listing; the evidence itself — dataset cards, rights summaries, privacy assessments, manifests — ships in evaluation packs and buyer diligence.
Dataset provenance methodology
Every dataset and program documents origin, purpose, consent basis and custody at collection. Provenance travels with the data: each delivery includes a manifest recording what was collected, under which specification, by which network class, and under which rights basis.
Chain-of-custody controls
From capture through processing to delivery, custody events are recorded and versioned. Deliverables ship with versioned manifests and checksums so a buyer can verify integrity and reconcile exactly what was received against what was accepted.
Rights-review process
Rights review precedes activation — a hard gate with no commercial override. The review covers the consent mechanism, authority to license, counterpart and bystander rights, and permitted model uses. Programs that cannot clear do not run.
Consent standards
Participants opt in at task acceptance with the purpose declared. This is materially stronger than retrofitting AI-training rights onto data gathered under general platform terms. Compensation is built into the participant payment flow.
Jurisdiction-specific privacy review
Privacy requirements differ by jurisdiction of collection. Each program is reviewed in the jurisdiction where collection occurs, covering local data-protection law, cross-border transfer requirements, and any all-party consent rules before any collection begins.
PII, biometric, voice, image and location handling
The capture specification defines the treatment of faces, plates, voices, precise location traces and any biometric identifiers: excluded, de-identified, or retained under a documented rights basis, per the buyer requirement and the jurisdiction review.
De-identification and deletion
Where de-identification is specified, the method (for example blurring, voice treatment, coordinate coarsening) is documented in the dataset card, with known limitations stated. Deletion and retention obligations are defined per program and honored across custody.
Quality assurance and independent adjudication
Two-stage verification: automated integrity checks plus adjudication independent of anyone who benefits from the outcome, against acceptance criteria agreed in writing before activation. Submitted and accepted volumes are reported separately, always.
Security, access control, retention and incident response
Data in custody is encrypted in transit and at rest, with access limited to program roles. Retention follows the program agreement. Security incidents affecting buyer data are disclosed to affected parties without undue delay. Security contact: daniel@kivotos.ai.
Secure delivery options
Delivery via secure transfer to buyer-controlled storage, with versioned manifests, checksums, consent attestation and rejection analysis accompanying every delivery.
Audit rights and documentation under NDA
Kivotos abstracts source identity commercially. Source-level documentation, network authority and audit rights are available to qualified buyers under NDA where a compliance obligation requires them.
Verification labels
A label appears on a listing only when the evidence behind it exists. Each label below states what it requires and the review tier applied. Current listings are collection capabilities: labels are applied per program as evidence is produced, and each label states whether verification was self-attested, internally reviewed, counsel-reviewed, or independently audited.
Origin verified
Collection origin and methodology documented and internally reviewed against the capture specification.
Rights reviewed
Rights basis, authority to license and permitted uses reviewed before activation; evidence available in diligence.
Human-generated verified
Composition reviewed for non-synthetic origin; any AI-assisted processing disclosed in the dataset card.
Privacy reviewed
Jurisdiction-specific privacy review completed for the collection context; assessment included in the evaluation pack.
Buyer-ready
Full documentation set available: dataset card, data dictionary, provenance summary, quality report, delivery sheet.
Kivotos does not use unsupported claims. You will not see “fully compliant”, “100% consented”, “zero PII”, or “guaranteed model improvement” anywhere on this site — those statements cannot be evidenced, and language that cannot be evidenced has no place in diligence.
Regulatory context
EU AI Act, Article 10 requires certain high-risk AI systems to address data origin and the relevant geographical, contextual and behavioural conditions of deployment. Kivotos documentation — origin, consent, custody, and collection context — is structured so buyers can meet provenance obligations with evidence rather than representations.