Signet Research publishes the research of Signet Systems. The projects study how AI systems handle controlled information: what a model can disclose or infer from what it was trained on, and how to measure whether access controls hold.
How a project is organized
Each project has its own page with up to five parts:
- Overview. The question, the current status, and how to cite the work.
- Explainer. A technical explanation of the method, written to be read without the paper.
- Paper. The abstract and the PDF. Drafts are labeled as drafts.
- Sandbox. An interactive version of the data or the scoring that runs in your browser. Nothing you enter is sent anywhere.
- Code. The repository and data, or what will be released and when.
Status labels
- Pre-registration draft. The design, hypotheses and analysis plan are published before the experiments run. There are no results yet.
- Preprint. Results are available. The paper has not been peer reviewed.
- Published. The paper has been peer reviewed. The page links to the published version.
Contact
Signet Systems can be reached through signet-systems.com.