Papers/2608.27512
🧪 Test?View on arXiv

Quantization-Triggered Backdoors in Language Models: Cross-Quantizer Transferability and the Validation--Deployment Gap

Not provided in the abstract

quantizationbackdoor attackslanguage modelsbehavioral equivalence
2608.27512
Builder Relevance
80%
2h ago

Abstract

The paper discusses the risks of quantization in language models, highlighting a validation-deployment gap that can lead to backdoor attacks.

Reality Card

Core Claim

The study proves that quantization can trigger backdoor attacks in language models, demonstrating that source-precision certification does not ensure behavioral equivalence in deployed configurations.

Method / Result

Backdoored translation models showed up to 85.02% inversion after quantization.

Limitations

The study's findings may vary across different quantization schemes and model architectures, complicating reproducibility.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers