Proposed Guidelines for the Responsible Use of Explainable Machine Learning

Explainability alone doesn't guarantee safety: The paper warns that explainable ML can be misused as a "faulty safeguard for harmful black-boxes," a practice called "fairwashing." This means companies could use explanations to make biased or problematic AI systems appear trustworthy without actually fixing underlying problems.

Explainable ML serves multiple critical purposes: Beyond transparency, the paper identifies that XAI enables human learning from models, allows people to appeal automated decisions, supports regulatory compliance, and helps audit systems for security—making it essential infrastructure for responsible AI deployment, particularly in financial services.

Explainability can be weaponized: The paper highlights that malicious actors can exploit explanation methods for harmful purposes, including stealing proprietary models and extracting sensitive training data, suggesting that organizations need safeguards around how they share explanations.

High-stakes decisions require interpretable models: The paper argues for using inherently interpretable models (rather than just explaining black-box models) and pairing them with debugging and bias-testing methods specifically in "life- or mission-critical ML systems"—such as those affecting healthcare, criminal justice, or financial access.

Guidelines are needed now, not later: Since explainable ML is already widely implemented in commercial systems and mandated in some industries, the paper emphasizes that best-practice guidelines must be established immediately to prevent harmful uses of this already-deployed technology.