Why biological AI needs its own safety rules

Biological artificial intelligence is not just a buzzword. It involves real-world use of AI models — from predicting protein structures to designing new drugs to simulating cellular aging. Models like DeepMind's AlphaFold have already shown that AI can solve in hours what previously took years. But with growing capabilities comes risk.

The problem lies in so-called black box models — AI systems whose internal decision-making process is non-transparent. When such a model designs a new molecule or genetic modification, it may not be clear at all whether the result is safe. "Traditional safety checklists fail here because there's no historical data to rely on," explains Justin Croyle, Chief Product Officer at Fennec Engineering. "When AI operates in uncharted territory, safety must be built from first principles."

Who is behind the partnership

The Buck Institute for Research on Aging is based in Novato, California, and is the only independent U.S. research institution focused exclusively on the connection between aging and chronic diseases. Their goal is not just to extend life, but primarily to extend healthspan — the years lived in good health. The project is led by Dr. James Yurkovich, who heads the Laboratory for Applied Systems BioAI at the Buck Institute (Yurkovich Lab).

On the other side is Fennec Engineering from Columbus, Ohio. The company has roots in Amazon Robotics and specializes in functional safety for autonomous systems. Its ASAP (Automated Safety Assurance Platform) already underpins the certification of more than a million robotic systems worldwide. Fennec's clients include Amazon, Meta, General Motors, and TÜV Rheinland. Now they are transferring their industrial robotics know-how into biomedical research.

How the safety system will work

At the core of the collaboration is an integrated safety framework that embeds automated safety checks directly into AI-driven biological simulations. These checks are not just an additional filter at the end of the process — they are part of the system architecture itself.

A key concept the partners are championing is called proactive trust. In practice, this means the system can predict and neutralize emerging risks before they leave the digital simulation environment. In other words — dangerous outputs should be captured and eliminated in the "lab incubator" of computer simulation, not at the stage of laboratory testing on living cells.

"As we incorporate more AI into our biological simulation research, it is essential to build the necessary safety guardrails," says Dr. Yurkovich. "This partnership allows us to move at the pace this emerging field demands, with confidence that safety is the number one priority."

From industrial robots to cellular simulations

Transferring experience from the industrial environment to biomedicine is not accidental. Functional safety is a discipline that says: a system must function correctly even when something fails. In autonomous vehicles, this means the car safely stops even when a sensor fails. In biological AI, it means that a model designing a new protein must be able to verify that its output will not cause unwanted cellular interactions.

Fennec Engineering brings the Build, Test, Certify methodology, already tested on over a million robotic systems. In the context of biological AI, this framework will be adapted to account for the specifics of biological systems — their complexity, feedback loops, and emergent behavior not found in mechanical systems.

What this means for aging research and medicine

The practical impact is enormous. AI models already help identify new therapeutic targets for treating Alzheimer's disease, cancer, or diabetes — diseases whose number one risk factor is precisely age. The safety infrastructure Buck and Fennec are building could dramatically shorten the path from computer simulation to clinical testing.

For the European and Czech context, this is important for several reasons. First, the EU AI Act classifies AI systems in healthcare as high-risk and requires demonstrable safety. The framework emerging from this American collaboration could serve as a reference model for European regulatory bodies as well. Second, the Czech Republic has a strong tradition in biomedical research — from the Institute of Organic Chemistry and Biochemistry of the Czech Academy of Sciences to biotech incubators in Brno — and similar safety standards will become increasingly relevant for Czech research teams.

Safety as a competitive advantage

The partnership comes at a time when investment in AI in healthcare is skyrocketing. According to a Yahoo Finance analysis, the AI in medicine market could reach a value of $3.36 trillion by 2035. In this context, safety infrastructure is not just a regulatory necessity — it is a competitive advantage. Research teams that can demonstrate their AI outputs are safe from the ground up will have an edge in drug approvals and grant acquisition.

With this partnership, Fennec and the Buck Institute are defining a new category — a safety supply chain for biological AI. Just as AWS cloud infrastructure enabled startups to scale without building their own server farms, this initiative could enable research teams to develop safe AI applications without building their own safety frameworks from scratch.