The AI Bioweapons Risk: We Cannot Afford to Look Away

AI creating and posing a bioweapons threat has become a documented finding, now on record. As such, health leaders and the wider biotech industry can no longer treat the AI bioweapon risk as someone else's regulatory problem.

A Microsoft-led team investigating biosecurity found that AI-redesigned toxin variants slipped past the genetic screening software that the DNA synthesis industry relies on to block dangerous orders.

The supply chains, the AI design tools, and the same synthetic biology platforms are all widely used across the entirety of pharma.

Image

Posing As a “Bioterrorist”

Researchers at Microsoft, working with academic and industry biosecurity specialists, used generative protein-design tools, including the company's own EvoDiff model, to redesign known deadly toxins such as ricin, botulinum, and Shiga toxin.

The goal was not to create functional bioweapons but to test whether AI-generated variants could be ordered from commercial DNA vendors without triggering the screening software those vendors use.

Across roughly seventy-two toxins and three open-source AI packages, this “red-teaming” exercise generated tens of thousands of protein variants and tested them against widely used screening tools.

A significant share evaded detection, prompting the researchers to treat the result as the biological equivalent of a zero-day vulnerability, alerting the U.S. government and screening-software vendors before publication so that patches could be issued ahead of public disclosure.

Crucially, the exercise was conducted entirely in silico. No toxic proteins were physically synthesised, and the researchers deliberately withheld the specific design techniques that enabled evasion.

The finding that matters for pharma biosecurity is that a safeguard the industry has long treated as a reliable backstop is no longer accurately catching what AI tools can generate.

Image

AI Biosecurity is a Governance Issue: Stop “Made-to-Order” Bioweapons

Pharmaceutical and biotech organisations sit directly inside the affected pipeline and therefore will play a vital role in protecting against the threat AI poses to biosecurity.

Commercial gene-synthesis vendors serve not only drug discovery, but also vaccine development and cell and gene therapy programmes across the industry. The same protein-design and generative biology tools driving therapeutic innovation are structurally identical to those used in the Microsoft study.

Writing in Nature, biosecurity researchers Jaspreet Pannu, Anita Cicero, and Tom Inglesby of the Johns Hopkins Center for Health Security argue that AI-enabled research could cause significant harm if used to design pathogens with concerning new properties.

Closing the gap, they argue, requires regulation imposed on the life sciences sector from outside and a closer collaboration between governments, AI developers, and biosafety experts.

This has teeth for procurement and compliance teams specifically. If screening gaps sit at the DNA-synthesis chokepoint, then vendor due diligence, not just internal research governance, becomes a frontline control.

Biosecurity Modernisation

United States lawmakers introduced the Biosecurity Modernisation and Innovation Act of 2026 (S.3741) in January 2026, sponsored by Senators Tom Cotton and Amy Klobuchar.

The bill would direct the Department of Commerce to mandate nucleic acid synthesis screening industry-wide, replacing the voluntary framework that has governed the sector for more than a decade.

Existing federal guidance already requires screened DNA for federally funded research, following the 2023 Executive Order on AI, but coverage outside that funding stream remains inconsistent.

A meaningful share of the DNA synthesis market does not screen orders at all, underscoring why a voluntary baseline struggles to hold across a fragmented global vendor landscape.

In parallel, the National Academies of Sciences, Engineering, and Medicine published a 2025 consensus study assessing how AI-enabled biological design tools intersect with pandemic-scale biosecurity risk.

They too recommended closer coordination between AI developers and biosecurity authorities alongside continued investment in next-generation screening technology.

Altogether, policy and technical mitigation now have to move together.

AI Can Be Used to Create a Bioterrorism Threat

AI's capacity to accelerate drug discovery and novel protein design is the same capacity that, misused, could lower the barrier to developing biological agents with greater precision and potency.

As a result, there has been a strong industry and intelligence call for layered technical safeguards alongside emerging policy frameworks, rather than relying on any single control.

The AI tools relevant to biosecurity can be distinguished into two groups: Large language models, which can broaden access to existing biological knowledge, and biological design tools, which can generate genuinely novel sequences of DNA and structures of proteins.

The convergence of the two, rather than either alone, is what most expands both the capability and the accessibility of biological misuse of AI over time.

None of the available research suggests AI-enabled biological attacks are imminent. It should be cautioned that disproportionate focus on screening software risks distracting from larger structural threats, including state-level biological weapons programmes, and it's confirmed that misuse attempts remain exceptionally rare.

The consistent message about the AI biosecurity risk is that the barrier is lower than it used to be with the advancement of AI capabilities in DNA and protein synthesis.

Image

What Pharma Leaders Should Do Now to Fend Against the AI Biosecurity Threat

Three actions translate directly into pharma board-level governance actions to protect against the AI biothreat risk:

1. Audit gene-synthesis vendor relationships. Confirm which suppliers screen orders against current sequence-of-concern databases, and require attestation as a procurement condition, not an assumption.
2. Treat AI design tools as dual-use from day one. Internal biosafety committees reviewing AI-assisted discovery programmes should explicitly assess misuse potential alongside therapeutic value, in line with the layered-safeguard approach the academic literature recommends.
3. Track S.3741 and its successors closely. Mandatory federal screening would reshape vendor qualification requirements industry-wide. Organisations that prepare compliance pathways ahead of enactment will face a far smoother transition than those that wait.

Generative Protein Design Poses Potential Biosecurity Threat

AI has not really made biological weapons easy to build; however, what research highlights is that it has made one specific safeguard, nucleic acid synthesis screening, measurably less reliable than the industry assumed, and pharmaceutical organisations sit close enough to that chokepoint that the gap deserves board-level attention rather than a footnote in the risk register.

At Pharmatica, we focus on the systems, strategies, and technologies shaping the future of pharmaceutical R&D, regulatory strategy, and manufacturing resilience. Our analysis sits at the intersection of policy, science, and operational reality, where governance decisions translate into measurable protection for patients, programmes, and reputations.

Pharmatica: Insight. Connection. Impact.