TraceTarnish tool uses Unicode injection to evade AI stylometry

Researchers have published an ablation study demonstrating that injecting zero-width Unicode characters and homoglyphs into text effectively neutralizes current stylometric authorship attribution systems. The paper, titled 'Occluded Oculus,' evaluates a framework named TraceTarnish designed to anonymize text against surveillance apparatuses by comparing translation, obfuscation, imitation, and injection modules. Experimental results indicate that character-level injection is significantly more effective than semantic rewriting at confounding multi-eyed stylometric classifiers. The authors frame this adversarial capability as a necessary privacy countermeasure against pervasive digital surveillance rather than a malicious exploit. This development highlights a growing technical asymmetry between text generation obfuscation and forensic detection capabilities. Security researchers warn that such tools could undermine efforts to attribute state-sponsored disinformation or verify anonymous sources in sensitive investigations. The methodology provides a reproducible blueprint for evading AI-based identity verification across commercial and governmental platforms.