Research question
How should third-party apps in LLM ecosystems be isolated from each other?
Key findings
- Hub-and-spoke architecture isolating apps with mediated communication - Prevents cross-app data leakage and hijacking at modest overhead
Applied operating-system isolation principles to LLM app ecosystems such as GPTs, published at NDSS 2025.
How should third-party apps in LLM ecosystems be isolated from each other?
Spot an error, or know a record we should add? Members can propose additions and submit corrections.