Global AI rules are built on a 'myth of transparency' even though making AI truly transparent is technically impossible.
March 19, 2026
Original Paper
Against Transparency
SSRN · 6433918
The Takeaway
Current laws like the EU AI Act assume that 'black box' algorithms can be made visible through documentation and disclosure. This paper argues that this is a conceptual mismatch with how machine learning works, suggesting that most current AI safety policies are chasing a goal that cannot be achieved by any existing technology.
From the abstract
Contemporary regulation of artificial intelligence is dominated by a powerful intuition: that the risks posed by opaque machine learning systems can be mitigated through transparency. This Article challenges that intuition. Drawing on both legal and computer science literature on explainable AI, it argues that transparency is not only technologically unrealistic for most modern AI systems but also conceptually mismatched with the kinds of accountability that law actually demands. What law requir