Saving Face: Investigating the Ethical Concerns of Facial Recognition Auditing

Auditing itself can cause harm: The paper demonstrates that well-intentioned efforts to audit facial recognition systems for bias can inadvertently harm the populations these audits are meant to protect, particularly because facial recognition data is highly sensitive and the technology is often deployed in ethically problematic contexts.

Five specific ethical tensions exist in facial recognition audits: The researchers identify a set of five distinct ethical concerns that auditors must navigate when evaluating commercial facial processing technology, suggesting that the auditing process involves complex tradeoffs that aren't always obvious.

Audits may complement existing harms: Rather than simply revealing problems, audits risk reinforcing or amplifying the harms already caused by biased facial recognition systems, particularly for vulnerable populations whose data may be misused.

Auditors need expanded design considerations: The paper calls for auditors to think beyond traditional performance metrics and consider broader ethical implications of their work, including how audit methodologies themselves might be misused or cause secondary harms.

Audits reveal fundamental product limitations: The research suggests that ethical concerns in auditing point to deeper, potentially unfixable problems with facial recognition technology itself—raising questions about whether some systems should exist at all, regardless of audit results.