Panel: The AI Architecture Debt—Refactoring Infrastructure for Sustainable Privacy
summary
Panel argues that AI systems accrue concrete architecture debt when models and RAG pipelines are wired without data provenance logging or the ability to honor deletion and portability requests, forcing expensive retrains or model deletions. Practical fixes include modular data boundaries, explicit privacy-aware model interfaces, automated red‑teaming and benchmarks tied to application contexts, and an architectural kill switch plus replayable logs for incident reconstruction. Examples cited: KBcache-style agent leaks, benchmarks like HELM being insufficient, and tens of thousands of detected issues that teams cannot yet triage.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:53AI architecture debt in privacy
- 6:07Privacy by design in AI
- 12:25Measuring AI debt
- 19:57Architectural shortcuts creating debt
- 22:48Refactoring for privacy
- 26:19AI governance distance
- 30:14Closing remarks
related talks
33mPanel: How Privacy Engineers Can Shape the Coming Wave of AI Governance
34mAccountability in the Age of AI Agents
01h 01mSoftware Architecture for Gen AI
11mHow Does AI Change Privacy? Towards Helping Practitioners Scope, Motivate, and Conduct
01h 06mDatabases and AI Training
50mPanel Software Engineering in the age of AI
39mBuilding a Responsible AI Governance Program
01h 02mArchitecture Patterns for AI-Powered Applications