Evidence with a chain of custody.
Every useful finding connects a source, product scope, point in time, and reviewer conclusion.
A clearer view of the public record.
Solid: recorded disclosure. Dotted: a link is on file but content was not extracted. Thin: not on file, which is inconclusive. These counts measure disclosure, not security or system behavior.
The register covers 6,895 domains, including 200 listed AI companies, and 3,805 published subprocessor relationships. 125 records have all four disclosure signals on file.
Source snapshot 28 Oct 2026, 8:54 AM PT. Counts describe published disclosures, not security or compliance. How to read the index
Discovery is the starting point
The public index records links to vendor trust centers and related artifacts. An indexed link is not proof that an artifact was inspected or that a control works.
Mappings need a reason
A meaningful mapping includes the framework and version, control ID, supporting source passage, product and service scope, rationale, reviewer, and review date. This first version suggests control topics. Exact framework mappings require researcher review.
Changes are observations
Monitoring preserves successful source versions, compares normalized content, and records changes. Content changes require materiality review. Access failures retain the previous evidence and produce an unavailable state.
Disclosure score for the 90+ shortlist
The public disclosure score gives 25 points each for a recorded trust or security page, named marks, a DPA URL, and named subprocessors. A page behind a recorded access wall or a subprocessor-list URL without imported names earns 15 points for that field. A page alone earns no mark points. Founding year is excluded. A 90 means three full signals plus a partial page or list; in this snapshot all 90s have a list URL without imported names. Source links and named marks have not necessarily been reviewed. This score measures public discoverability, not security effectiveness, data exposure, breach history, or overall trustworthiness.
Aggregate filer scores rate public files
The existing public-file score sums five record fields worth 0, 10, or 20 points each: a recorded public page, named marks, a DPA URL, named subprocessors, and founding-year metadata. It does not establish that a certification is current, a DPA was inspected, or a control works. Founding year is metadata rather than a security disclosure. For a published subprocessor name, each distinct vendor contributes at most one score. The median is primary and the mean is secondary; missing scores are excluded. Because inclusion requires a filed subprocessor list, the aggregate is structurally weighted upward. It does not score the subprocessor, measure risk or security, or establish legal identity.
Verification has a scope
A researcher may confirm an artifact, its applicability, or a control within a specified evidence period. These are different conclusions. A general vendor trust score cannot replace them.
Research and verification powered by SecurityPal AI
The offering combines automated collection with researchers and partners. In this preview, review requests are stored as drafts and are not dispatched to a live research team.
Read the public record.
Browse published trust pages, security standards, and source links. Missing evidence means unknown, not unsafe.
Explore the 90+ shortlistFollow the subprocessor list.
See which vendors name the same providers. Trace each relationship back to its published source.
Explore the graphKeep your work moving.
Save vendors, follow profile updates, and connect your agents in one private workspace. Start free.
Evidence you can point to.
Public disclosure is a starting point. OpenTrust separates what a vendor publishes from researcher review and control verification.