Disclosure and context
A task can communicate more than its individual fields. The research question is how context, combinations and repeated interactions affect what another party may learn.
OIA LAB · RESEARCH
Understand the problem.
Explore the questions.
Our research program examines how task context can disclose sensitive information and how defined execution boundaries can be evaluated.
TERM INTRODUCED JANUARY 2026 · RESEARCH NOTE
Cognitive leakage: a definition
What an external AI can infer about your intent from the task you give it, even when every sensitive field has been redacted. How it differs from data leakage, where it comes from, and how intent isolation responds. OIA Lab introduced the term in January 2026 and has been researching it since.
Read the noteEvaluation methods
Useful evaluation separates directly assigned information, observed outputs, reconstruction attempts and task quality. Each measurement depends on its fixture and assumptions.
Execution records and review
An execution record can support review within a defined collection and verification scope. It should not be confused with proof of every downstream event or inference.
READ EACH PUBLICATION IN CONTEXT
A research result is not automatically a product result.
Check the publication date, version, methods and limits. Broader OIA research provides background; it should not be read as a test of the current APL implementation unless the work explicitly evaluates that version.
START WITH A NON-CONFIDENTIAL DESCRIPTION
Bring a research or evaluation question.
Tell us which question or capability you are evaluating. Start with a non-confidential description so we can discuss the relevant research and its scope.
Discuss the researchPlease do not send client documents, deal details or unpublished invention material through the website.