Morning Edition · Monday, August 31, 2026Published at 2:23 AM EDT · New York

New Preprint Shows Quantization Can Switch On Hidden Backdoors That Full-Precision Testing Never Sees

Backdoored translation models measured clean at 16-bit precision produced corrupted output in up to 85.02 percent of cases after routine post-training compression.

A preprint posted to arXiv today challenges an assumption that most deployment pipelines rely on without stating it: that post-training quantization is a semantically neutral optimization. The authors argue that because quantization maps ma…