Least-privilege access
Reviewers should see only the projects, items, and fields required to do the work.
Security for Expert AI Data Programs
Security for expert AI data programs is mostly access design: who can see which items, in which tool, for how long. This page describes operating principles. It does not claim SOC 2, HIPAA, ISO, GDPR certification, or any other audit badge.
We are not a freelancer marketplace or traditional staffing company.
What security means on an expert AI data program
Security is mostly access design: who can see which items, in which tool, for how long. Programs can use least-privilege, project-scoped access, confidentiality terms, data minimization, and revocation when work ends. This page does not claim SOC 2, HIPAA, ISO, or GDPR certification.
Client-controlled tools where applicable
When the safer design is to keep data in the client environment, experts work there. Labels do not need to travel to be useful.
Project-specific access
Access is granted for a named engagement, not as a standing right to a general data lake.
Confidentiality agreements
Experts are contracted with confidentiality obligations appropriate to the project.
Data minimization
If a reviewer does not need a patient identifier, a customer name, or a live deal, that field should not be in the item.
Access removal after completion
When the project ends, access is intended to be revoked rather than left idle.
Secure document handling
Downloads, screenshots, and local copies are treated as design choices to avoid, not as a convenience default.
Auditability where supported
If the client environment can log who saw which item, that log is part of the operating design. No universal audit product is claimed.
Project-specific requirements
Contact us to discuss project-specific security requirements. The correct design for a public benchmark is not the correct design for unpublished clinical notes or live financial documents. Those differences should be settled before the first reviewer is onboarded.
Discuss security constraints before staffing
Include the data class, the required environment, and any hard geographic limits in the first brief.
- Need 25 licensed nurses for a clinical AI evaluation
- Need 15 attorneys in a specific practice area
- Need 20 PhD scientists for benchmark creation
- Need 30 senior software engineers for code evaluation