Rethinking Insurance and Legal Responsibility for AI-Enabled Biology

After a car accident or house fire, two questions immediately follow: who was responsible and who pays for the damage? The answer lies in liability and insurance, and these systems shape our behavior well before anything goes wrong.

People and companies who know they can be held responsible have reasons to take precaution. Insurers can encourage safer practices through what they charge and the conditions of coverage. How the costs of an accident are allocated can influence how hard people work to prevent it.

Questions about AI companies’ responsibility for harm are already reaching the courts. In January 2026, Character.AI and Google agreed to settle a lawsuit alleging that a chatbot contributed to a teenager’s death. In March, a father sued Google and Alphabet, alleging that Gemini contributed to his son’s death by suicide; Google said the chatbot was designed to discourage self-harm and directed him to crisis support.

In September, British Columbia OpenAI, alleging that the company failed to alert law enforcement to threats made on its platform before the Tumbler Ridge mass shooting.

These cases raise broader questions about how responsibility is shared across companies, what precautions they should take, and when they might bear responsibility for harm.

As AI expands what is possible in biology, it’s past time the biosecurity, insurance, and legal communities came together to examine what these developments mean for preventing harm and assigning responsibility if it occurs.

Old Liability Questions; New Territory

AI is helping researchers design proteins, investigate disease, and develop medicines⁠, while also raising concerns about accidents and deliberate misuse. The fact that this work can draw on tools and services from several companies—with one company supplying the AI model, another making the biological material, and a third may run the experiment—further complicates the issue.

That division of labor can help research move faster while making a familiar question hard to answer: who is in the best position to prevent a bad outcome, and who should pay if one happens?

Biology, with its natural ability to replicate and spread, adds another problem by enabling harm across borders and over time. These characteristics raise questions about how familiar insurance and liability arrangements might apply, and make addressing this increasingly important.

Why AIxBio Risks Are Different

Consider the difference between contamination that temporarily closes a laboratory and an infectious release that spreads beyond it. Both can be costly, but an infectious release may continue causing harm without the laboratory doing anything further. Its reach depends partly on how quickly it is detected and contained. Understanding the risk means considering both what could cause an incident and what could allow its consequences grow.

An incident can also start in one country and harm people in several others. Some losses may not surface immediately with a clear causal connection. Which country’s courts decide the case? Which insurance policies pays? What does an organization owe people it will never meet, in places where it will never operate? These are difficult questions to settle after any incident, and they are ones we should be asking today.

Many Puzzle Pieces; Unclear Responsibility

Assigning responsibility gets harder when several organizations share the work. The AI developer knows what its model can do. The researcher knows the purpose of a project. The laboratory controls the experiment. Each holds a piece of the puzzle, but none sees the whole picture. Their individual decisions shape the outcome, so understanding those contributions has to come before drawing conclusions about liability or insurance.

Anthropic’s September 2026 threat report⁠ illustrates one aspect of this problem. A platform serving life-sciences researchers took requests that Claude had refused and routed them to other companies’ models with more permissive safeguards. Anthropic did not assert that the researchers intended harm, but the case shows how an intermediary can change what users can access across providers. Judging one company’s safety practices tells you only part of the story when services are combined. That matters most when the workflow proceeds from a screen to a laboratory bench.

Earlier crises offer a reason to begin this conversation before a serious incident. When COVID-19 closed businesses in 2020, many did not know if their business interruption policies would pay. The United Kingdom’s Financial Conduct Authority took a test case to court on their behalf, and the Supreme Court settled the main questions in January 2021. Businesses and insurers needed answers while the crisis was unfolding. Although AI-enabled biology will raise different questions, that experience shows the importance of timing. Unresolved insurance questions get much harder to answer when people are already counting their losses.

Building the Conversation for Tomorrow’s Risks

As biosecurity practitioners, we are beginning to explore what these developments mean for insurance and liability, and how these systems might encourage the responsible use of AI in biology.

We can help explain how biological harm might occur. We cannot tell you where today’s policies and legal doctrines work, where they fall short, or what might need to change. The answers may look different for accidents and deliberate misuse, or for losses confined to one organization and those affecting many others.

We invite colleagues in insurance, law, industry, and public policy to join this conversation. Bring your experience with emerging risks and the questions you think we should be asking.

Our goal is to convene a group to explore these questions and develop practical steps toward addressing them, including how insurance and legal responsibility could prevent harm. If you would like to be involved, please contact Daniel Irowa-Omoregie at [email protected]