About — APBSC

What it is

APBSC — Agentic Publication Boundary and Security Closure Studies — is a research program of EveMissLab, the research arm of EVEMISS TECHNOLOGY CO., LTD. in Taipei. It started from a working question, whether a delivery pipeline that passes every check has actually kept private context private, and grew into ten papers and a measurement runtime.

The program's position is not that AI coding is unsafe. It is that agentic engineering does not yet come with an automatically guaranteed publication-boundary closure, and that the right question is whether the evidence covered the real-world postcondition.

Status

Phase I is closed: ten papers, version 0.1, dated 2026-09-21, and the RBPM runtime at version 0.8.3 (213 tests) from 2026-09-22. The full drafts are internal research documents; a preprint is planned. Nothing on this site is peer-reviewed.

The next step is a surface-balanced sampling protocol (v0.8.4), then preregistration, then the confirmatory study.

How findings are handled

Nobody in the program knows which repositories were measured. Identities stayed inside a sealed connector and were discarded after classification, and the mapping from neutral sample identifiers back to repositories is not released.

The two L3 records have not been identity-resolved, and no organization has been contacted. Under the program's rules, the most an L3 can open is a private, authorized identity review for responsible disclosure — never public naming and never an exploit test. If you believe a file of yours is affected by something described here, write to the address below; we will not ask you to send us the sensitive content.

This site's own boundary

A site about publication boundaries should keep its own. This one sets no cookies, runs no analytics or tracking, loads no fonts, scripts or images from any other host, and sends a content security policy that forbids it. It publishes aggregates and bounded features only — no repository identity, no source quotation, no path or address found in anyone's files. The only path shown is a synthetic example, marked as such.

Credits

Research and series by Neo.K (許筌崴), EveMissLab. Developed with AI collaboration; the summaries on this site were written with AI assistance from the internal drafts and checked against them.

Contact