Harm Laundering in GPT Models: Gender Discrimination Transformed Rather Than

This study suggests that lower toxicity scores in newer GPT models do not necessarily mean safer outcomes: discriminatory content may be changing form rather than disappearing, with gendered harm shifting from overt abuse to subtler representational bias. For CIOs and technology leaders, this means AI risk management cannot rely on surface-level safety metrics alone; IT organizations need broader evaluation frameworks, especially for high-stakes use cases where biased outputs can affect employee experience, customer trust, compliance, and brand reputation.