On AI Safety and Security Technical Debt in Engineering AI-Enabled Systems

AI systems accumulate hidden "technical debt" across multiple areas — including data management, model design, and testing — that differs from traditional software debt because these problems often remain invisible until they cause failures in high-stakes domains like healthcare and autonomous driving.

Seven major sources of AI problems have been identified: data governance issues, model implementation flaws, algorithm design choices, architectural decisions, operational processes, documentation gaps, and inadequate testing. These interconnected problems can cascade through AI pipelines, making systems harder to maintain and less reliable over time.

AI safety and security risks are deeply intertwined with trust: The research maps 31 types of technical debt to 18 trust-related concerns, including 6 specific safety hazards and 12 security vulnerabilities, showing that responsible AI requires addressing both dimensions simultaneously rather than treating them separately.

Practical mitigation strategies exist across the AI lifecycle: The study synthesizes 34 actionable guidelines (8 focused on safety, 26 on security) for preventing, detecting, and reducing these problems, suggesting that organizations can take concrete steps to build more trustworthy systems rather than accepting these risks as inevitable.

Proactive management is essential because problems hide in plain sight: Unlike traditional software bugs, many AI technical debts are latent and only emerge under certain conditions, making systematic documentation, testing, and governance practices critical for catching issues before they affect real people.