arXiv:2511.19299v1 Announce Type: new 
Abstract: Novel deep learning architectures are increasingly being applied to biological data, including genetic sequences. These models, referred to as genomic language mod- els (gLMs), have demonstrated impressive predictive and generative capabilities, raising concerns that such models may also enable misuse, for instance via the generation of genomes for human-infecting viruses. These concerns have catalyzed calls for risk mitigation measures. The de facto mitigation of choice is filtering of pretraining data (i.e., removing viral genomic sequences from training datasets) in order to limit gLM performance on virus-related tasks. However, it is not currently known how robust this approach is for securing open-source models that can be fine-tuned using sensitive pathogen data. Here, we evaluate a state-of-the-art gLM, Evo 2, and perform fine-tuning using sequences from 110 harmful human-infecting viruses to assess the rescue of misuse-relevant predictive capabilities. The fine- tuned model exhibited reduced perplexity on unseen viral sequences relative to 1) the pretrained model and 2) a version fine-tuned on bacteriophage sequences. The model fine-tuned on human-infecting viruses also identified immune escape variants from SARS-CoV-2 (achieving an AUROC of 0.6), despite having no expo- sure to SARS-CoV-2 sequences during fine-tuning. This work demonstrates that data exclusion might be circumvented by fine-tuning approaches that can, to some degree, rescue misuse-relevant capabilities of gLMs. We highlight the need for safety frameworks for gLMs and outline further work needed on evaluations and mitigation measures to enable the safe deployment of gLMs.

أظهرت دراسة حديثة تقييمًا لصلابة نموذج اللغة الجينومية Evo 2 من خلال التعديل العدائي باستخدام تسلسلات من 110 فيروس ضار يصيب البشر. تبرز هذه الأبحاث المخاطر المحتملة المرتبطة بتطبيق هياكل التعلم العميق على البيانات البيولوجية، خاصة في توليد الجينومات لفيروسات قد تصيب البشر.

Un estudio reciente evaluó la robustez del modelo de lenguaje genómico Evo 2 mediante un ajuste adversarial utilizando secuencias de 110 virus dañinos que infectan a los humanos. Esta investigación destaca los riesgos potenciales asociados con la aplicación de arquitecturas de aprendizaje profundo a datos biológicos, especialmente en la generación de genomas para virus que podrían infectar a humanos.

Une étude récente a évalué la robustesse du modèle de langage génomique Evo 2 par le biais d'un ajustement adversarial utilisant des séquences de 110 virus infectant les humains. Cette recherche met en lumière les risques potentiels associés à l'application des architectures d'apprentissage profond aux données biologiques, en particulier dans la génération de génomes pour des virus pouvant infecter les humains.

A recent study evaluated the robustness of the genomic language model Evo 2 through adversarial fine-tuning using sequences from 110 harmful human-infecting viruses. This research highlights the potential risks associated with the application of deep learning architectures to biological data, particularly in generating genomes for viruses that could infect humans.

Open-weight genome language model safeguards: Assessing robustness via adversarial fine-tuning

Was this article worth reading? Share it

LucidQuery AI

Humanize AI

Airparser