K. Provatas

Adversarial protein sequences can hijack generative models, forcing them to produce specific, unintended protein designs with minimal poisoning data.

Model-generated DNA can now be traced with perfect accuracy, even after single-base edits, using a watermark detectable without knowing its precise location or strand.

The biomedical significance is that, under the stated threat model, a served genomic model can process an encoded sequence without exposing plaintext queryderived activations to the compute provider.