AI biology needs security built in, researchers say
Generative artificial intelligence is increasingly able to do more than analyze biological data. It can design DNA, RNA and protein sequences, capabilities that may support drug discovery and new therapies but also create opportunities for misuse. A research team led by University Medical Center Mainz is calling for security measures to be embedded throughout these systems from the outset.
The team’s recommendations appear in Nature Communications in a paper titled A defense-in-depth strategy for securing AI-driven biological sequence design. Their proposed approach treats safety as a layered technical and organizational problem rather than something that can be addressed with a single safeguard.
Design capabilities bring dual-use risks
Biological sequence-design systems can generate outputs relevant to proteins and viral genomes. That makes them potentially valuable to medicine and biotechnology, including efforts to develop drugs and therapies. But the same ability to generate biological designs can be dual use: tools intended for beneficial research could potentially be directed toward harmful ends.

The researchers point to risks that include the creation of new pandemics or pathogens affecting crops. The concern is not that every biological AI system will cause such outcomes, but that rapidly expanding design capabilities require protection mechanisms calibrated to the possibility of misuse.
A defense-in-depth model
The proposed response is a multilayered—or defense-in-depth—security architecture. Under this model, no individual control is expected to carry the whole burden. Instead, security measures would span the process chain, beginning with training data and extending through the development and use of AI models for biological design.
That framing matters because biological AI systems involve more than a final generated sequence. Data selection, model construction, user access and downstream handling can all shape risk. A layered approach aims to reduce reliance on any one point of failure and to make security part of the system’s design rather than an afterthought.
An interdisciplinary proposal
The recommendations came from a three-day ideas workshop held in November 2025 by the Wübben Stiftung Wissenschaft. Twenty experts from molecular biology, AI, bioinformatics, ethics, medicine and politics took part in developing strategies for responsible use of generative AI in biology.

Lead author Dr. Maximilian Sprang, a researcher at University Medical Center Mainz and group leader at Johannes Gutenberg University Mainz’s Institute for Quantitative and Computational Biosciences, described the underlying challenge as one of managing dual-use risks while preserving beneficial innovation.
Why the proposal matters
The research does not describe a ban on AI in biology. Instead, it makes the case for safeguards that match the technology’s growing ability to create biological sequences. That distinction is important: the same advances that may accelerate biomedical research can demand more deliberate security planning.
As AI becomes more deeply embedded in life-science workflows, questions about how models are trained, accessed and governed will become practical engineering and policy issues. The Mainz-led team’s central recommendation is straightforward: security should be integrated early and repeatedly across the AI biology pipeline, not added only once a system is already capable of producing sensitive designs.
This article is based on reporting by Medical Xpress. Read the original article.
Originally published on medicalxpress.com







