DNA sequence screening and integrated biosecurity controls are becoming core infrastructure for modern biopharma as AI‑accelerated design and accessible synthesis dramatically increase the volume and novelty of constructs being ordered and built. We interviewed Kevin, CEO of Aclid, on how biosecurity standards could shape and be shaped by the industry. This article will feature the highlights from our interview and information about emerging guidances and how to build a robust biorisk management system that can strengthen biopharma’s access to public funding, improve partnering with large pharmaceutical companies, and maintaining public trust in biotechnology.
What Is Biosecurity?
In biopharma, biosecurity refers to the policies, technical controls, and organizational practices that prevent the loss, theft, misuse, or deliberate release of harmful biological materials, information, and capabilities. It complements biosafety (which focuses on protecting workers and the environment from accidental exposure) by focusing on intentional or unauthorized misuse, dual‑use risks, and access control to valuable biological materials such as pathogens, toxins, engineered constructs, and enabling technologies. For drug developers, CDMOs, and DNA/RNA providers, biological security encompasses secure sequence and customer screening, robust biorisk management in labs and manufacturing, governance of AI‑enabled design tools, and alignment with evolving national and international biosecurity guidelines so that powerful platform technologies can be scaled without creating unacceptable security risks
Why biosecurity practices matter for DNA sequence screening
DNA and RNA synthesis have become dramatically cheaper and more accessible, while AI-driven design tools now generate vast numbers of novel constructs that no human can manually review at scale. This combination increases both the upside of synthetic biology and the downside risk that dangerous or poorly understood sequences are accidentally or deliberately synthesized, including highly modified variants that do not closely resemble known pathogens. Robust biosecurity screening—tied to well‑curated “sequences of concern” lists and modern DNA synthesis guidance such as the HHS/OSTP frameworks—is therefore a critical control point: it is the last common checkpoint before designs become physical molecules and enter laboratories, supply chains, and ultimately clinical or industrial environments.
Q: How does Aclid’s platform complement sequence screening workflows?
A: We work with much of the DNA synthesis industry to build DNA synthesis screening tools that help identify potential hazards in DNA, RNA, and protein orders. We screen against a wide range of lists, including regulated lists such as export control regimes, the U.S. Federal Select Agent Program, and Schedule 5 in the UK, as well as biosafety and biosecurity guideline lists from agencies like the NIH and their counterparts in countries such as Germany, which include oncogenes and other non‑infectious disease risk factors.
Our system aggregates these lists, allows customers to choose which sets they want to screen against, and is tested via red‑teaming exercises and industry benchmarks. We have collaborated with groups such as Microsoft and major DNA synthesis companies like Twist and IDT to evaluate performance on complex, obfuscated, and AI‑generated sequences using advanced screening algorithms and sequence alignment techniques to ensure resilience in detecting highly modified constructs and pathogenic sequences.
Q: What is the sentiment in the industry right now among companies creating or assembling DNA? While some show limited interest in screening, Ribbon Bio appears to be an exception. How does this affect your work?
A: The International Gene Synthesis Consortium (IGSC), which Ribbon Bio is an active member of, is a core industry group that shares best practices, follows screening guidelines, and works to advance standards and, ultimately, regulation. In some cases regulators and policymakers request information from the IGSC, and the group provides input on how DNA synthesis screening is being implemented in practice.
Note: DNA synthesis screening standards and best practices are not only relevant for DNA providers, but also across the biotechnology ecosystem, including clinical laboratories and diagnostics stakeholders.
Some DNA synthesis providers consider screening critical and have invested in it even without regulatory mandates, driven by liability concerns and a belief that it is essential for a maturing market. Others have only basic or rudimentary protocols, often because they are dedicating fewer resources. Most companies are not opposed to screening in principle; rather, screening requires focused attention, staffing, and operational capacity. The IGSC, policymakers, and governments are working toward a level playing field in which everyone follows common baseline standards.
Q: How does regulation affect a pharma company’s choice of DNA provider? If they select a provider that does not screen, what risks could that pose when it is time to file, for example, for a Biologics License Application (BLA)?
A: There are two main risk dimensions: access to funding and internal safety/liability. On funding, agencies such as the U.S. NIH have updated recombinant DNA guidelines and funding policies to state that funded entities should screen the DNA they order, with associated IBC processes and protocol review. Similar expectations increasingly spill over to other U.S. agencies (DOD, BARDA, NSF) and international funders that look to NIH as a standards‑setter. Choosing a synthetic DNA provider that does not screen could jeopardize eligibility for federal funding or certain strategic partnerships.
The second risk is to employees and operations. Modern R&D increasingly uses AI and high‑throughput design tools to generate large numbers of constructs, far beyond what a human can manually review. This creates “bio error” risks, where dangerous or inappropriate sequences are created inadvertently. Without robust screening for biological agents and pathogen identification, such sequences could be synthesized and shipped into labs, creating safety and biosecurity risks, particularly for groups working with pathogens or infectious disease models.
What you need to know about bio-error
Bio error refers to harm or risk arising from unintentional mistakes in biological design. It matters because AI and automation amplify oversight errors that can reach labs or patients unless strong screening and governance layers are in place.
Q: Given that there are not yet binding biosecurity regulations, mostly guidance, how far should a DNA provider go in safeguards and screening so that when regulations arrive, their existing standards can still be relied on?
A: Today, there is a combination of industry and governmental standards that effectively define the “forward‑compatible” baseline. On the industry side, the IGSC Harmonized Screening Protocol lays out guidelines for sequence and customer screening. On the government side, there are U.S. Department of Health and Human Services screening guidelines, UK guidance, relevant OSTP framework elements, and applicable ISO standards for compliance.
While these are not yet mandatory regulations, aligning with them builds a strong foundation for future laws currently moving through legislatures, such as U.S. federal bills referencing these same standards, the proposed EU Biotech Act, and emerging UK policies. U.S. states like California are also advancing screening requirements with associated fines. Providers that already comply with IGSC standards and HHS guidance on nucleic acid biosecurity will be in a much stronger position to adapt quickly and avoid liability when these frameworks become enforceable regulation.
Note: There are numerous considerations in deciding to follow any one guidance. As these regulations are developing, there are different perceptions of risk in the industry. One major concern is delaying shipment of DNA and losing business to competitor companies. But when biosecurity automation is integrated in manufacturing and resource planning systems, this does not have to be the case. At Ribbon Bio, we believe that building robust biosecurity practices are not a burden but rather an investment that improves our DNA, safeguards customers and partners, and strengthens the industry.
Q: You mentioned more automated risk assessment. What are the gaps today in how risk is assessed, and how do you see this evolving over the next five to ten years?
A: Current best practice combines automated customer risk profiling with automated sequence screening. On the customer side, tools integrate KYC, sanctions checks, and identity verification using public and private data, plus automated follow‑ups to fill information gaps. On the technical side, sequence‑screening systems flag hazardous sequences. Together, these create a holistic view that allows 99% of orders to pass without manual review while focusing resources on the small fraction that are flagged, with additional tooling to pre‑populate compliance forms for import, export, or transport licenses.
The main gap is functional assessment, especially as AI‑generated constructs become common and sequences diverge further from naturally occurring references. Even today, designed constructs may have no close natural analogues, making function hard to infer and causing false positives. Over the next five to ten years, we expect greater integration of function‑prediction models that can distinguish whether a flagged similarity relates to binding, cytotoxicity, catalysis, signaling, or other mechanisms, thereby reducing false positives and providing richer, function‑level insight to reviewers. Some of this already exists in background data curation; more will be surfaced directly in real time to customers.
Q: How important is interpretability in these AI models? How should providers think about this, given that everyone may have their own risk‑assessment models?
A: Interpretability is critical for trust, governance, and regulatory engagement. One approach is to require models to cite sources, so outputs are grounded in specific literature or databases rather than free‑floating speculation. This creates guardrails and lets reviewers inspect the underlying evidence used in risk assessments.
In biology, additional interpretability comes from examining internal representations of models, such as embeddings from protein language models like ESM. By comparing vector representations of sequences to reference sets, one can identify regions of structural or functional similarity, then map those to experimental structures and domain annotations. Even if the model itself is not transparent, this post‑hoc analysis highlights regions and mechanisms worth deeper investigation. In practice, these models guide reviewers toward the most relevant experimental data and structural information, rather than serving as black‑box oracles.
Q: Are there other AI model types—such as large protein models—you use or would recommend, beyond large language models?
A: Yes. Not all biological problems require transformer‑based large language models. Many tasks can be addressed with other architectures, including recurrent neural networks and generative adversarial networks (GANs). For example, GANs can be used where one model generates sequences and another discriminates between real and generated examples, which can be useful for specific design or detection tasks.
Transformer architectures are particularly powerful when supervised data are sparse and the goal is broad representation learning or reasoning over diverse sequence spaces. However, in many biological domains we have rich supervised datasets and detailed mechanistic understanding, making narrower, task‑specific models more appropriate and efficient. In those cases, existing architectures and targeted supervised learning can outperform generic, very large models, and not everything should be replaced by a frontier LLM‑style system.
Q: Thinking about protein engineering, integrating AI into a design–build–test–learn (DBTL) workflow poses challenges. What do you see a fully automated DBTL workflow looking like when biosecurity checks are integrated throughout?
A: “In a highly automated DBTL workflow, error management is essential across quality, efficacy, and biosecurity. On the quality side, companies already use tox‑screening models and developability filters to rule out molecules that will never be manufacturable or safe. On the discovery side, lead‑finding and optimization models limit exploration to promising regions of sequence space.
Biosecurity needs analogous layers. Screening should be continuous during design, flagging candidates that drift into unsafe or disallowed sequence space before they are ever ordered. Then, before build, sequence screening at the synthesis stage—whether from commercial DNA synthesis technology or benchtop DNA synthesis systems—acts as another gate. The goal is a layered “Swiss cheese” model, where each screening layer is imperfect on its own but overlapping layers reduce the chance that a dangerous or unusable construct passes through the full DBTL cycle. Integrating security and safety models at both design and build stages is key to making automated workflows robust rather than brittle.
What a robust biosecurity foundation would look like
Organizations should define clear escalation and documentation pathways for flagged orders, routinely red‑team their screening systems, and coordinate with industry consortia and regulators so that their internal controls are aligned with external expectations and standards. Effective risk mitigation requires screening both double-stranded DNA and single-stranded DNA constructs, along with comprehensive biosafety protocols. The goal is to build multiple, mutually reinforcing layers of defense—at design, ordering, synthesis, and in the lab—so that if one control misses a problem, others are still likely to detect it before a risky construct is produced or used.
Q: What do you see as the largest looming threats in biosecurity over the next five to ten years, as AI becomes more integrated into DNA assembly and pharma workflows?
A: The biggest risk may be societal and regulatory backlash if biosecurity is not managed proactively. Biology is intrinsically higher‑stakes than domains like cyber, yet cyber enjoyed a long period of relatively unmanaged risk before robust standards and infrastructure were built. A similar pattern in biotechnology—major incidents before guardrails are in place—could trigger severe reactions.
If the synthetic genomics industry were perceived as unsafe or irresponsible, we could see contraction in investment, abandonment of promising research areas, stricter bans, or broad public rejection of biologics or engineered products. To avoid that outcome, the sector must pair technical innovation with visible, credible safeguards and standards. Ensuring robust screening, responsible AI use, and strong safety practices is not just about compliance; it is a prerequisite for the long‑term growth, legitimacy, and social license of the entire bio‑innovation ecosystem.
Protecting the social license for biotechnology
Beyond technical risk, many observers argue that the long‑term viability of synthetic biology depends on maintaining public trust and a “social license” to operate. Analyses of biotech governance emphasize that transparent, science‑based risk assessment and visible safeguards are essential: when standards are well‑constructed and clearly communicated, transparency tends to increase trust; when they are weak or opaque, attention can reduce trust and fuel calls for over‑correction and restrictive bans. In the context of DNA synthesis and AI‑accelerated design, commentators warn that gaps in screening and safety mechanisms risk high‑profile incidents that could trigger political backlash, investment pullback, or blanket restrictions on otherwise beneficial technologies. Proposals therefore stress voluntary pre‑market testing, harmonized screening norms (e.g., via IGSC, HHS/OSTP, ISO), and proactive engagement with regulators and civil society as ways to show that the industry is capable of governing itself responsibly, rather than waiting for crisis‑driven regulation.
