Having access to information doesn't give anyone the right to transform it, test it or license it. Algorithm7 treats each of those as a separate permission, confirms it before the work begins, and carries it into everything we produce.
That protects the organizations who contribute data, and gives the labs who evaluate and license it a provenance record their legal teams can review.
Source ownership, processing authority, evaluation permission and training rights are separate questions. Each activity is authorized in writing and recorded before it happens. If it isn't authorized, it doesn't happen.
As information moves through all seven stages of Algorithm 7, its rights move with it. A restriction set on the source record doesn't disappear when the record becomes a task, a training example or a delivery. Select a restriction to follow it through each stage.
Buyers may test derived tasks, but nothing may be used for training until a license grants it.
Removing names doesn't guarantee anonymity; combinations of details can still identify someone. So we assess re-identification risk per dataset and combine several safeguards, rather than relying on any one.
Names and direct identifiers
Contact details and addresses
Credentials and secrets
Pricing and commercial terms
Confidential IP outside scope
LocationsRegion bands
DatesRelative intervals
PeopleRoles
Rare attributesCoarsened or dropped
ValuesSynthetic, where needed
Re-identification risk assessment
Expert review for sensitive data
Access controls on every product
Contractual limits on re-identification
Leakage checks before release
We don't promise that buyers can never encounter sensitive information unless processing, review and access controls can enforce it. Handling is set by sensitivity, not convenience.
Standard
Operational and technical data with low personal content after conditioning.
Rights-cleared samples may be shared
Standard conditioning and review
Licensed delivery to approved buyers
Elevated
Data with personal, financial or commercially sensitive content.
Evaluation through blind benchmark or API grading
Expert re-identification review
Delivery under strict access terms
Restricted
Health, clinical or other regulated data.
Restricted processing environment
Additional consent or legal authority as required
Controlled evaluation only, no downloadable sample
Expert review before any release
Removing identifiers doesn't override an NDA, satisfy a consent requirement or create a licensing right. Authorization comes first. Protection is applied on top of it, never instead of it.
What a data owner authorizes sets the outer limit. Evaluation terms and buyer licenses can only ever sit inside it.
What the data owner authorizes
Data categories and ingestion
Permitted processing, including AI models
Derived assets and sample disclosure
Evaluation and licensing scope
Compensation, retention and deletion
What a buyer may do while testing
Qualified buyers only
Evaluation environment and access
No training on evaluation materials
What may be retained afterwards
Confidentiality of results
What a buyer may do after licensing
Defined permitted uses
Supplier restrictions inherited
Sublicensing and derived outputs
Provenance documented per delivery
Conditioned products, never raw originals