Permissions for coding-assistant search Personal proposal by Per-Arne Andersen S7 · Spring 2027 https://thesis.uya.no/proposals/an-assistant-that-knows-its-boundaries/ Prevent a coding assistant from retrieving documents the current user is not allowed to read. TECHNICAL - Build top-5 source retrieval with permission filtering before ranking and default denial. - Compare filtering after global top-5 selection without refilling, using the same policy and ranker. DATA (planned) 30 fictional documents, 3 roles, 60 queries Ten documents per project. Define membership and copied-content rules; label authorized relevant IDs for all 180 role/query pairs, including empty-access cases. METHOD Measure unauthorized text/IDs/citations and authorized-document recall. Project-only filtering is an additional negative control. Four IT users interpret denials and escalation options; score comprehension against the written policy. OUTPUT Access-controlled search, permission fixtures and leakage report. BACHELOR Task: Build document search that filters results using a fixed role-permission matrix. Data: 30 fictional documents, 3 roles, 60 queries (planned) Ten documents per project. Define membership and copied-content rules; label authorized relevant IDs for all 180 role/query pairs, including empty-access cases. Requires: IT participants and independent policy review; synthetic findings do not establish enterprise readiness. Method: Use the fictional repository documents and role-labelled queries; test allowed retrieval and prohibited content/identifier leakage. Plan 4 participant sessions. Output: A permission-aware search demo and exhaustive policy tests. Use relevant literature to justify the established approach; a new research contribution is not the aim of this proposal. MASTER Investigate where permission filtering should occur in a retrieval pipeline. Compare pre-filtering and post-filtering with the same role policy; jointly evaluate authorized recall, leakage and whether users understand denials. Intended contribution: Evidence about security–retrieval trade-offs and usable denial explanations. STARTING PAPERS Gao et al. (2024 revision) — Retrieval-Augmented Generation: A Survey: https://arxiv.org/abs/2312.10997 Separate retrieval, evidence use and output evaluation; trace newer retrieval work. Nissenbaum (2004) — Privacy as Contextual Integrity: https://digitalcommons.law.uw.edu/wlr/vol79/iss1/10/ Analyse recipient, purpose and information flow rather than treating privacy as a role label. Search Scopus or Web of Science and ACM Digital Library using the topic query, then follow citations to the thesis start date. Record searches and compare methods, data, findings and limitations in literature-matrix.csv. Use that review to confirm or revise the gap and choose a current comparator. The linked papers are starting points. START HERE Tools: Python, BM25 library, Streamlit, pytest 1. Write 3 fictional documents with intentionally different document-level permissions. 2. Create the full role/document policy matrix and expected permitted results before implementing retrieval. 3. Test filtering before ranking on the pilot documents; separate denied retrievals from answer-generation claims. Literature search: AI assistant access control information disclosure user understanding Study controls: - Determine allowed/denied balance separately for each role, not globally for a query. - Primary comparison keeps the same policy; coarse project-only filtering is explicitly a weaker-policy negative control. - Omit answer generation. Secure retrieval does not establish generated-answer confidentiality. - Before collecting participant data, agree consent, storage and withdrawal handling with the supervisor. Use participant codes, not names, in study files. - For the master’s study, use the research task above to define the factors and comparisons in this pilot plan. Preregister one primary outcome and feasible scope after the literature review; do not add every possible model or interface variant. REQUIRES 4 IT participants and independent policy review; synthetic findings do not establish enterprise readiness. -------------------- NON-TECHNICAL - Prepare assistant answers showing sources drawn from three fictional projects. - Ask users which sources each role should be allowed to see. DATA (planned) 12 paper access scenarios + 8 IT interviews 8 determinate and 4 ambiguous cases covering roles, copied information and cross-project answers. Prepare equivalent prose policies and permission tables. METHOD Compare policy comprehension for determinate cases with balanced, disjoint versions. Analyse ambiguous ownership cases and escalation reasons separately; do not score ambiguity as a definite error. OUTPUT Policy clarification table and source-display recommendations. BACHELOR Task: Document who should access each kind of coding-assistant document. Data: 12 paper access scenarios (planned) 8 determinate and 4 ambiguous cases covering roles, copied information and cross-project answers. Prepare equivalent prose policies and permission tables. Requires: Recruit software/IT staff; no assistant implementation. Method: Use the paper role/document scenarios; record determinate decisions separately from policy ambiguities. Plan 4 participant sessions. Output: A permission matrix and an escalation path for unclear requests. Use relevant literature to justify the established approach; a new research contribution is not the aim of this proposal. MASTER Explain mismatches between permission policy and users’ expectations of assistant access. Contrast prose and permission-table cases using contextual integrity; separate ambiguous policy from incorrect interpretation. Intended contribution: A theory-based account of access expectations and unresolved governance decisions. STARTING PAPERS Nissenbaum (2004) — Privacy as Contextual Integrity: https://digitalcommons.law.uw.edu/wlr/vol79/iss1/10/ Analyse recipient, purpose and information flow rather than treating privacy as a role label. Orlikowski & Gash (1994) — Technological Frames: https://dl.acm.org/doi/10.1145/196734.196745 Compare how roles interpret the purpose, operation and use of the same system. Search Scopus or Web of Science and ACM Digital Library using the topic query, then follow citations to the thesis start date. Record searches and compare methods, data, findings and limitations in literature-matrix.csv. Use that review to confirm or revise the gap and choose a current comparator. The linked papers are starting points. START HERE Tools: LibreOffice Writer/Calc, audio recorder with consent; no programming required. 1. Prepare a pilot with 2 examples from: 12 access scenarios + 8 interviews. Write the task questions and a reference answer sheet. 2. Write a recruitment message, information sheet and consent form for the participants named above. Agree privacy handling with the supervisor before contact. 3. Pilot one session after approval; revise unclear questions, freeze the task sets and coding categories, then recruit the planned sample. Literature search: AI assistant access control information disclosure user understanding qualitative scenario study Study controls: - Independently check policy labels and define how copying affects a document’s classification before recruitment. - Before collecting participant data, agree consent, storage and withdrawal handling with the supervisor. Use participant codes, not names, in study files. - Pilot separately, then freeze the questions and coding plan. Check objective answer keys independently; keep an audit trail of coding, including disagreements. - For comparisons, counterbalance order and case assignment; do not show a person both versions of one case. Report participant-level findings, not repeated tasks as independent people. REQUIRES Recruit software/IT staff; no assistant implementation. Bachelor: apply established methods and evaluate a practical solution or study. Master: position a research question in current scientific literature, investigate a mechanism or unresolved problem, and explain the contribution. Final scope is agreed with me. Study sizes are proposed; participant recruitment and planned materials are not already arranged.