AIxBio Horizon Scan Finds Accelerating Capabilities Require Urgent Oversight

The convergence of artificial intelligence and the life sciences continues to accelerate. The AIxBio Horizon Scan: Spring 2026—the second in a series published under the AIxBio Global Forum—shows that while individual advances may seem incremental, their combined trajectory is significant. Progress is outpacing governance frameworks.

What’s Changed Since Winter

Since our first Horizon Scan, protein design tools have continued to improve, with new models consolidating capabilities into faster, more accessible platforms. Agentic coding tools—AI systems that can write, execute, and debug code with minimal human intervention—have matured, making computational biology workflows available to a much wider group of researchers. AI companies are making substantial investments in the life sciences, signaling that biology is becoming a core commercial frontier for AI development.

Frontier AI models released in April also showed meaningful gains on biological benchmarks. Both Anthropic and OpenAI have launched purpose-built life science products, marking a shift from general-purpose AI applied to biology toward dedicated scientific tooling. This builds on trends we flagged in the winter scan while concentrating advanced capabilities behind managed-access enterprise programs.

A Widening Gap Between Capability and Oversight

Several developments underscore the urgent need for stronger governance. Data—long identified as a bottleneck for biological AI tools—is increasingly becoming proprietary. Companies are building competitive advantages around closed datasets, which could concentrate the most powerful capabilities among a small number of well-resourced actors and potentially reduce visibility into those capabilities. This trend expands beyond the United States; China’s Shanghai Zhangjiang science hub is integrating AI-driven molecular design with synthesis and testing under one roof, further accelerating global development.

Meanwhile, new empirical studies are beginning to assess whether AI actually meaningfully assists individuals with real-world biological work. Early results are nuanced: AI does not yet significantly help novices perform hands-on laboratory tasks, but it does provide substantial uplift for computational biology. These studies are far more informative than standard benchmarks for understanding actual biosecurity risk, and more are needed.

Policy Progress and Its Limits

The most notable policy development is the introduction of the bipartisan Biosecurity Modernization and Innovation Act (S.3741) in the United States. This bill would establish mandatory federal requirements for DNA synthesis providers to screen orders and verify customer identities—a welcome step that builds on years of advocacy by NTI and partners to move beyond the current voluntary framework.

However, analysts note that the bill’s reliance on sequence-matching approaches may be insufficient. AI design tools could, in the near future, generate functional biological agents with novel sequences that would not match existing watchlists, making complementary, function-based screening essential.

Geopolitically, reductions in U.S. government research funding are beginning to take effect, and the gap between leading AI programs globally has narrowed. Track II dialogues on biosecurity and AI safety remain an important channel for communication between American and Chinese experts, but recent quotes from U.S. Treasury Secretary Scott Bessent following the May 14-15 U.S.-China summit also signal potential Track I cooperation on these issues.

Looking Ahead

The trajectories identified in our first horizon scan are largely continuing as expected. Capability development is still outpacing governance, and the window for establishing effective oversight mechanisms continues to narrow. In the coming months, we anticipate further integration of AI agents with laboratory systems, new generations of protein and genomic language models, and the emergence of closed-loop discovery systems capable of producing compounding capability gains that are both difficult to anticipate and hard to reverse.

NTI | bio will continue to publish Horizon Scans every six months to support well-informed decision-making by policymakers, industry leaders, and researchers. Getting ahead of these risks requires understanding what’s coming—and acting before the window closes.