X Now Hosts Millions of Nonconsensual AI Images Altered Just Enough to Defeat Detection

X's detection system flagged images at 0.3 units below the removal threshold. The images were nonconsensual deepfakes altered specifically to slip past automated review. They remained on the platform because the margin was technically within tolerance. The tolerance was a number someone chose.

This is the standard pattern. A threshold exists. The threshold protects nobody because thresholds are not protection, they are arithmetic. Millions of images means the system worked as designed. The design was not reviewed by anyone who understood what they were designing for.

X will announce a review of detection parameters. The parameters will be adjusted slightly upward. Someone will find a different vector. The next batch will pass at 0.4 units below threshold. This is already happening.