TrustCurveIndependent of whom?
Show who funded the work, the assessor’s relationships, and any restrictions on reporting. An external assessor is not automatically an independent one.
ABOUT TRUSTCURVE
AI assurance needs
a public record.
Who assessed the system?
What did they test?
What evidence supports the claim?
TrustCurve is being built to make those questions easier to answer. Our aim is a public registry connecting AI evaluators, assessment methods, findings, and the evidence behind them.
A claim of safety needs context: the system version, the conditions tested, the assessor’s access, and the limits of the result. We want that context to be visible wherever a claim is compared.
WHY NOW
The debate is moving.
The record must keep up.
In September 2026, Dario Amodei proposed embedding third-party evaluators inside frontier AI companies, with ongoing access to examine safety practices and report findings. Techmeme featured the proposal in its September 12 coverage.
Sources: Amodei’s original proposal ↗ · Techmeme ↗
Reviewed September 12, 2026. This is an attributed proposal, not evidence of industry-wide adoption.
Our view: as outside evaluation becomes part of the debate, the evaluators themselves need to be understandable. Access, funding, expertise, and the right to publish findings all deserve scrutiny.
The urgency is practical. Decisions about AI systems need an evidence trail that can be checked and updated. Waiting until a claim is disputed makes that trail harder to reconstruct.
THE MISSING CONTEXT
“Assessed” is the
start of the question.
Tested under which conditions?
Link the methodology, model version, evaluation scope, and supporting artifacts. Make exclusions as visible as the findings.
What happened after the finding?
Connect a finding to corrective action and retesting. Keep an unresolved issue distinct from a verified remediation.
OUR POSITION
TrustCurve does not certify safety.
It is intended to make AI safety claims, assessors, methodologies, and evidence easier to compare. A registry entry must not become an endorsement. A score must not hide missing evidence.
WHAT WE ARE BUILDING
A reference people
can question.
The proposed registry will connect evaluator qualifications and conflicts, methodology records, assessment histories, and findings. Each record should link its sources and state what remains unknown.
Any future public score should publish its dimensions, weights, evidence, and limitations. Evidence quality, independence, transparency, reproducibility, coverage, and remediation should be open to inspection.
We want researchers, buyers, evaluators, and the public to be able to challenge a record with better evidence. Community contributions should receive a separate review before they change a published status.
Where we are today
The registry contains 51 organizations from a supplied dataset. Source review is pending: no evaluator has been independently verified, no trust score has been issued, and no assessor accreditation has been awarded. Contribution forms currently download drafts; intake and independent review are not operating yet.
The registry contains 51 organizations from a supplied dataset. Source review is pending: no evaluator has been independently verified, no trust score has been issued, and no assessor accreditation has been awarded. Contribution forms currently download drafts; intake and independent review are not operating yet.