The Central Commission on Biological Safety (ZKBS), represented by the Synthetic Biology Working Group, is addressing the rapid advances in the use of artificial intelligence to generate genetically modified organisms (GMOs). This encompasses the de novo generation of nucleic acid sequences and proteins facilitated by bioinformatic methods of machine learning and generative artificial intelligence in silico, along with their synthesis and further development through self-reinforcing learning loops in automated laboratories. The ZKBS strongly appreciates this research, as it has the potential to contribute, among other things, to innovative development pathways leading to novel products and to a better understanding of evolutionary mechanisms. Furthermore, the newly developed methods can also provide valuable support for the risk assessment of genetic engineering work.
The ZKBS considers the article titled “Semantic design of functional de novo genes from a genomic language model” [1], published in Nature in 2025, to be representative of developments in this research field.
This publication deals with the generation of novel prokaryotic genes using a large language model (LLM) trained on a dataset comprising approximately 80,000 prokaryotic genomes and two million phage genomes. Through rational design, this LLM enables the generation of proteins based on the semantic syntax and evolutionary conservation of prokaryotic gene clusters and operons. Building on knowledge of the functions of conserved proteins within these clusters and operons, it is possible to generate genes with desired properties and functions in silico by interacting with the LLM in a targeted way. The authors demonstrate that novel functional proteins can be produced from these in silico-generated genes using bioinformatic filtering steps. Their synthesis can be based either on chemical or genetic engineering.
As a result, the de novo generated nucleic acid sequences or proteins are comparable to the products of classical mutagenesis or approaches of directed evolution, but they may also be so novel (new to nature) that they cannot be attributed to any donor organism.
When assessing the hazard potential of genetic engineering operations involving the production of GMOs, the hazard potential of the donor organism must be taken into account in accordance with the provisions of the German Genetic Engineering Safety Ordinance (GenTSV). From the ZKBS’s perspective, the risk assessment of GMOs produced using new to nature nucleic acid sequences is also possible under current genetic engineering law even if these nucleic acid sequences cannot be attributed to a donor organism (see § 4, sentence 1, No. 1(d) GenTSV). The risk assessment is conducted under § 5 GenTSV in conjunction with Annex 1, whereby particular consideration is given to the known and predictable information regarding the transferred nucleic acid sequence (see § 5(1)(2) GenTSV in conjunction with Annex 1 No. 2.1(b)). Accordingly, the GMO may also be assigned to a higher risk group if it cannot be ruled out that a transferred nucleic acid sequence increases the GMO’s hazard potential.
published June 2026