AI governance for space biotech: risks, funding and operational steps

L’artificial intelligence enters today into laboratories and orbits and asks to be governed. This article brings together three concrete cases to explain what it means for an Italian founder or investor. It analyzes public funding to a space biotech, the ability of a genomic model to create functioning viruses, and the organizational transformation that turns tools into operational practices. The goal is practical: indicate risks, opportunities and operational choices to take immediately.

Risks and opportunities must be managed with concrete operational measures.

L’artificial intelligence enters today into laboratories and orbits and asks to be governed.

Public money, experiments in orbit and what Orbital High-Throughput Screening measures

Exobiosphere is a Luxembourgish biotech startup founded in 2024 by Kyle Acierno, Bruno Santos and Olivia Borgue. It received €1,000,000, that is the maximum cap of the Jeune Entreprise Innovante program. This non-dilutive public aid cofinances up to 70% of the financial needs and operates under the jurisdiction of Luxembourg. Luxinnovation is the agency that certifies the innovative company status required to participate in the program.

The product at the center of the investment is called Orbital High-Throughput Screening. The platform is designed to run over 2,000 automated experiments simultaneously per mission. Control is mirrored on Earth in real time. The platform feeds and differentiates cells, doses compounds, processes samples autonomously and returns plate readings and microscopic images. This means complex experimental data can quickly reach research teams and pharmaceutical partners.

Complex experimental data can quickly reach partners.

The choice to use microgravity derives from its ability to accelerate processes that on Earth occur slowly. Microgravity allows observing cellular senescence, immune dysfunction and tumor aggressiveness in a different way. For a founder this entails a dual requirement. Space engineering skills and solid data governance practices are needed. If data become a product, clear roles for biosafety and human checkpoints before any synthesis or launch are required. Human checkpoints are required before synthesis.

The model that created 16 viruses: technical fact, practical implications and gaps in screening

Evo 2 is a genomic model developed by researchers at Stanford University and the Arc Institute. The model has 4 billion parameters and was trained on 9.3 trillion nucleotides. Its context window reaches up to 1,000,000 base pairs. It generated almost 700,000 candidate bacteriophage genomes. Among about 300 genomes synthesized and tested, 16 proved viable in the laboratory. These phages replicated inside antibiotic-resistant Escherichia coli cells. Genomic models can produce functional sequences in the laboratory.

These results open a fast path to designing therapeutic phages against resistant strains. The researchers made Evo 2 freely available, including parameters and code. Biosafety experts at Johns Hopkins University warned that current synthetic DNA order screening systems are not equipped to detect sequences generated in this way. In many jurisdictions there is no obligation requiring DNA suppliers to perform extensive checks.

Current DNA screening standards are insufficient.

For a startup that mixes generative models with experimental automation the operational question becomes simple and urgent. It is necessary to decide where to insert the “human in the loop”. One must define who reviews and authorizes critical steps, from design to synthesis. Procedures for control with DNA synthesis suppliers must be established. These procedures must go beyond current standards.

Companies must implement human checkpoints and advanced synthesis controls.

Why artificial intelligence requires a different organization, not just new tools

The difference between giving access to tools and rethinking work is crucial. The common approach, defined here as AI-sprinkle, yields limited gains. Teams that receive tools achieve improvements around 30 percent. To obtain multiple-fold gains it is necessary to adopt AI-native practices. Job descriptions, routines and workflows must be redefined. In this way artificial intelligence generates hypotheses and people remain orchestrators and final decision-makers. AI-native practices transform how teams work and decide.

AI-native practices transform productivity gains.

An emblematic case is HireRoad, which had planned to rewrite a product in 18 months. The company chose to redesign the organization and workdays with an AI-native mindset. It completed the reconstruction in 15 weeks, therefore much earlier than the original plan. The first 34 customers have already migrated and the team became smaller and more senior. Key actions were operational training, rapid prototyping with customer feedback and processes in which the model proposes solutions presented to a human for final judgment.

For a space biotech these practices must integrate biosafety. Roles that authorize sequence synthesis and protocols for incidents in orbit must be defined. Clear lines of responsibility are required. Without reorganization, technical speed risks becoming a vulnerability.

Reorganization is necessary to make technical speed safe.

Prospettive e prossimi passi

The three stories converge on operational and policy decisions. Evo 2 shows that it is possible to generate 16 functioning viruses in the laboratory. Exobiosphere obtained €1,000,000 to take experiments into orbit. The organizational experience shows that turning a 30 percent gain into exponential increase requires revolution, not tweaking. Founders and investors must translate these truths into concrete choices. They must appoint senior figures for biosafety and codify human checkpoints. They must negotiate advanced screening clauses with DNA suppliers. Founders must appoint senior biosafety figures immediately.

Operational choices must be made immediately.

The regulatory knot remains open. Institutions will have to decide obligations for synthesis suppliers and rules for publishing generative models. Meanwhile companies can act now. They can create operational checklists, train responsible figures and adopt AI-native practices that include the “human in the loop”. In this way technical speed can be transformed into value without exposing the community to avoidable risks.

Sources:

• Luxembourg’s space BioTech Exobiosphere secures €1 million to find… | eu-startups.com
• L’AI ha creato 16 virus mai visti in natura. E sono funzionanti | ainews.it
• AI-Native, Not AI-Sprinkle: Why AI Is A Business Change, Not A… | news.crunchbase.com