False Sense of Security in Explainable Artificial Intelligence (XAI)

Explainability requirements in AI regulations may create a false sense of security: The paper argues that EU and US regulations (including the AI Act and GDPR) mandate explainability as a core requirement for compliant AI systems, but current XAI methods frequently produce "erroneous, misleading, and incomplete explanations," meaning compliance alone doesn't guarantee trustworthy AI.

No single XAI method is reliably superior: The research finds that among the many available explainability techniques, "none of which presents a clear edge," and each method has "multiple failure modes" requiring customized development and ongoing evaluation—suggesting that choosing an explanation method is not straightforward.

"Explainability" lacks a consistent definition: The paper identifies that "explainability" has multiple meanings often used interchangeably across policy and technical contexts, creating confusion about what regulators actually require and what companies should implement.

Current AI governance risks becoming performative compliance: Unless governments establish clearer legislative guidance grounded in technical realities, the paper warns that AI regulation risks becoming a "vacuous 'box-ticking' exercise" where organizations appear compliant without meaningfully addressing safety and accountability concerns.

Technical limitations undermine policy objectives: The fundamental disconnect between what regulations promise (trustworthy, accountable, transparent AI) and what current technology can deliver threatens the effectiveness of AI governance frameworks designed to protect the public.