Framework for Designing Fully Original, Functional De Novo DNA Sequences

Section One: Introduction

1.1 Research Background and Significance

Macro Societal & Disciplinary Context

For decades, life science operated in a read-only paradigm. Projects like the Human Genome Project spent billions and more than ten years merely mapping existing natural DNA sequences, while synthetic biologists relied on trial-and-error tinkering to rearrange pre-existing genetic material. Traditional genetic engineering could only cut, copy, and paste natural gene fragments; fully original, functional DNA with no evolutionary precedent remained out of reach. Starting in the early twenty-twenties, genomic foundation models upended this constraint, and bioengineer Eric Nguyen’s TED talk AI that designs entirely new DNA sequences centers on his landmark Evo model—a seven-billion-parameter genomic AI capable of generating full-length, biologically functional de novo DNA from zero evolutionary templates. Today, biotech, drug development, climate engineering, and space biology race to adopt generative genomic AI, yet the field lacks unified frameworks balancing transformative innovation with biosecurity and ethical guardrails. Nguyen’s presentation bridges lab technical breakthroughs with public discourse, unpacking how AI shifts biology from a science of discovery to a discipline of intentional creation.

Practical Significance

This analysis translates Nguyen’s Evo AI framework into actionable workflows for synthetic biologists, pharmaceutical developers, biomanufacturing engineers, biosecurity regulators, and policy designers. For lab researchers, it breaks down Evo’s end-to-end pipeline for generating novel CRISPR systems, phage genomes, and therapeutic gene circuits, cutting traditional multi-year design cycles down to weeks. For healthcare practitioners, it outlines how AI-generated DNA enables personalized gene therapies tailored to unique patient genetics. For global regulators, it identifies unmonitored biothreat risks from unrestricted de novo DNA generation and builds tiered oversight protocols aligned with Nguyen’s public risk warnings. For commercial biotech firms, it delivers cost-reduction benchmarks, as AI virtual screening eliminates more than ninety percent of costly failed wet-lab experiments.

Theoretical Significance

Classical molecular biology theory operates around the central dogma of naturally evolved DNA-RNA-protein flow, with no formal framework for predicting function in fully artificial genetic sequences. Nguyen’s Evo research fills this critical theoretical gap by establishing a unified multimodal genomic learning model that simultaneously interprets and generates DNA, RNA, and protein sequences without species-specific fine-tuning. Prior synthetic biology scholarship treated sequence design as modular rearrangement of natural parts; Nguyen’s work constructs a new generative theory of genetics that accounts for long-range genomic context, regulatory non-coding regions, and emergent biological function in AI-invented code. It also expands bioethics frameworks by introducing unaddressed questions about ownership, environmental release, and dual-use risks of human-designed non-natural life code.

1.2 Core Concept Definitions

1. De novo DNA sequence design: The creation of fully original genetic code with no direct evolutionary counterpart in natural organisms, generated computationally rather than copied or modified from existing plant, microbial, or animal DNA—Evo’s core capability as highlighted in Nguyen’s TED talk.
2. Genomic foundation model (Evo): Nguyen’s StripedHyena-based large AI architecture trained on two point seven million prokaryotic genomes, with a one hundred thirty-one kilobase context window that processes full genomic segments at single-nucleotide resolution to predict and generate functional DNA, RNA, and protein systems.
3. Design-build-test-learn (DBTL) closed loop: The iterative synthetic biology workflow Nguyen standardizes around Evo: AI sequence generation, chemical DNA synthesis, wet-lab functional validation, and model retraining using experimental failure/success data to refine future designs.
4. Dual-use genomic AI risk: The inherent danger that tools built to create therapeutic, environmental, or industrial DNA can be misused to generate novel pathogenic sequences, unregulated viral genomes, or toxic protein coding regions with no natural analogs.
5. Functional novelty threshold: The benchmark Nguyen establishes to distinguish trivial sequence variation from transformative de novo design: AI-generated DNA that performs biological tasks no known natural gene can replicate, such as custom CRISPR-Cas editing complexes or anti-resistance bacteriophages.

Clarification of Easily Confused Concepts

Many observers conflate AI-optimized natural genes with fully de novo DNA design. Nguyen draws a sharp dividing line: optimization tweaks existing natural sequences for better performance, while de novo generation writes entirely new base pair strings with no evolutionary lineage. A second widespread confusion mixes generic large language models (LLMs) with genomic foundation models; ChatGPT-style text AI cannot capture nucleotide folding, regulatory spacing, or protein translation rules, whereas Evo is pretrained exclusively on raw genomic data for biological coherence.

Scope and Boundaries

This analysis centers entirely on Eric Nguyen’s TED talk arguments and peer-reviewed Evo research, focusing on de novo DNA generation for microbial, therapeutic, and industrial biological applications. It excludes consumer-facing DNA editing tools, CRISPR tweaks to human patient germlines, and eukaryotic whole-genome design (still unvalidated at scale with Evo). The scope prioritizes technical workflow, industry use cases, and dual-use biosecurity ethics as Nguyen frames them in his public presentation.

1.3 Current Research Status, Domestic and International

Key Developmental Milestones

Before twenty twenty-three, genomic AI was limited to short protein coding fragments with context windows under one thousand base pairs, unable to model full gene circuits or transposon systems. NeurIPS twenty twenty-three released HyenaDNA, the precursor architecture to Nguyen’s Evo, unlocking long-range genome context processing. In twenty twenty-four, Nguyen’s team published landmark Science results proving Evo could design and lab-validate novel CRISPR complexes and full synthetic phage genomes—this work formed the technical backbone of his twenty twenty-five TED talk. Since then, global labs have replicated Evo’s core pipeline, yet uniform global oversight for generative DNA tools remains absent. National bodies like the U.S. NIH and EU ECHA have released scattered advisory guidelines, with no binding international treaty governing de novo sequence synthesis and AI access.

Mainstream Scholarly Viewpoints

Two competing frameworks dominate generative genomic AI research:

1. Unrestricted innovation camp: Most commercial synthetic biotech firms argue open access to Evo-style models accelerates medical and climate breakthroughs, framing biosecurity risks as manageable through post-design screening only.
2. Nguyen’s balanced governance framework (this article’s guiding lens): The interdisciplinary academic position presented in the TED talk, which advocates for pre-generation function screening, tiered model access limits, and mandatory wet-lab validation to mitigate dual-use harm while preserving legitimate research progress.

Shortcomings and Controversies in Existing Research & Industry Practice

Four unresolved gaps Nguyen highlights during his TED presentation persist across global biotech ecosystems:

1. Severe training data bias: Most genomic AI is trained on common lab microbes, lacking diverse eukaryotic, extremophile, and environmental organism data, limiting generalization for novel ecosystem design.
2. Black-box model opacity: Evo and competing genomic foundation models cannot fully explain why a specific de novo sequence achieves function, slowing targeted refinement and trust among clinical regulators.
3. Disconnected design-synthesis pipelines: Many research labs separate AI sequence generation from DNA synthesis vendors, creating unmonitored gaps where hazardous de novo code could move from digital design to physical material without review.
4. Uneven global access disparities: Wealthy nations and large biotech firms own nearly all high-capacity genomic AI hardware, while low-resource research communities lack access to safe screening and validation infrastructure, widening global biotech risk inequity.

1.4 Framework and Core Objectives

Article Logical Structure

This piece uses the Problem-Solution format (Module D), aligned with Nguyen’s TED talk structure: first cataloging transformative opportunities of AI de novo DNA design, then mapping interconnected technical, biosecurity, and ethical risks, analyzing layered root causes of unregulated generative genomics, citing Evo’s validated lab trials as a benchmark advanced model, and delivering tiered technical, institutional, and global policy safeguards. Later sections cover cross-industry real-world applications, pervasive public and lab misconceptions, actionable practitioner guidance, forward-looking research trajectories, and formal citations anchored to Nguyen’s TED presentation and peer-reviewed Evo publications.

Core Guiding Research Question

How does Eric Nguyen’s Evo genomic AI enable fully original, functional de novo DNA sequence design, what overlapping technical and dual-use risks emerge from unrestricted access to this technology, and what multi-layered lab, corporate, and international governance solutions balance generative genomic innovation with biosecurity as outlined in his TED talk?

Key Takeaways for All Readers

1. Evo’s long-range multimodal architecture eliminates the historic barrier preventing AI from writing complete, biologically functional non-natural DNA sequences, delivering custom genetic tools impossible to build via natural gene rearrangement.
2. Unregulated open access to de novo DNA generation creates unique dual-use risks: AI can invent novel pathogenic genetic circuits undetectable by traditional sequence similarity screening.
3. Nguyen’s standardized closed-loop DBTL workflow drastically cuts synthetic biology development timelines and failure rates, yet requires mandatory experimental validation to confirm AI-generated sequences perform as predicted.
4. Sustainable, safe scaling of generative genomic AI demands coordinated tiered access controls, pre-synthesis function screening, and binding international biotech governance, not voluntary industry self-regulation alone.

Section Two: Main Body (Option D — Problems and Countermeasures)

2D.1 Overview of Core Risks and Unmet Limitations of AI De Novo DNA Design

Nguyen’s TED talk balances celebration of Evo’s breakthrough with clear identification of four interconnected categories of systemic risk and technical limitation:

1. Unpredictable emergent biological function in untested de novo sequences AI generates DNA based on statistical genomic patterns, not full simulation of cellular environment interactions. Even well-trained models produce sequences with hidden off-target effects: unintended toxic protein expression, cross-species horizontal gene transfer, or disrupted host cell metabolism that only appear after physical synthesis. Without mandatory wet-lab validation, AI-designed DNA risks unplanned environmental or patient harm if deployed prematurely.
2. Unfiltered dual-use hazard generation bypassing traditional pathogen screening Conventional biothreat tools flag DNA matching known viruses or bacterial toxins. Evo can invent entirely novel pathogenic genetic cassettes with zero homology to any recorded organism, slipping past sequence-matching security filters. Malicious actors could leverage simplified genomic AI interfaces to design unrecognizable infectious agents without advanced molecular training.
3. Severe technical bottlenecks scaling from short genes to full complex genomes Evo reliably generates gene circuits, CRISPR systems, and small phage genomes up to six hundred fifty kilobases, yet struggles with large eukaryotic multi-chromosome designs. Long-range epigenetic regulation, non-coding enhancer networks, and cell-type specific expression remain poorly modeled, limiting near-term use for complex multicellular organism engineering.
4. Fragmented global oversight creating cross-border biosecurity loopholes No universal rules govern access to high-capacity genomic AI or commercial DNA synthesis order screening. A de novo hazardous sequence designed on an unregulated cloud AI platform in one country can be ordered from an overseas synthesis vendor with zero coordinated review, evading national safety laws entirely.

2D.2 In-Depth Root Cause Analysis

Four overlapping structural failures create the unregulated risk landscape Nguyen addresses in his TED presentation:

Root Cause One: Training Data and Model Opacity Limitations

Genomic foundation models rely on incomplete, biased public genome datasets missing rare environmental organisms. Simultaneously, deep learning’s black-box nature removes clear causal links between AI prompt and final DNA sequence, making pre-synthesis risk prediction incomplete without physical lab testing.

Root Cause Two: Decoupling of AI Design Tools and DNA Synthesis Regulation

Digital genomic AI platforms operate outside the scope of most biosecurity legislation, which historically only regulates physical DNA manufacturing facilities. Regulators lack jurisdiction over cloud-based sequence generation tools, creating a critical split where dangerous code can be designed digitally before oversight activates at the synthesis stage.

Root Cause Three: Commercial Incentives Prioritize Speed Over Risk Mitigation

Biotech startups and pharmaceutical firms compete to reduce development timelines, creating pressure to skip rigorous multi-stage functional screening for AI-generated sequences. Mandatory validation cycles add cost and delay product launches, so voluntary safety protocols are frequently deprioritized without regulatory enforcement.

Root Cause Four: Fragmented International Biotech Governance

Existing global biosecurity agreements (such as the Biological Weapons Convention) predate generative genomic AI and contain no provisions for de novo digital sequence oversight. National regulations vary wildly, enabling regulatory arbitrage where developers shift high-risk AI workflows to jurisdictions with minimal screening rules.

2D.3 Reference to Advanced Benchmark Experience: Eric Nguyen’s Evo Closed-Loop Lab Pipeline

Nguyen’s own research lab serves as the gold-standard model for balancing de novo DNA innovation and risk control, cited repeatedly throughout his TED talk:

1. Pre-generation functional constraint prompting: All Evo design requests require detailed documentation of intended biological function, host organism, and containment level; the model’s prompt filter blocks requests targeting vertebrate toxicity or transmissible pathogen traits before sequence generation begins.
2. In silico multi-layer risk screening: Post-AI generation, automated computational tools evaluate every candidate de novo sequence for emergent virulence factors, off-target binding, and horizontal transfer potential before synthesis approval.
3. Tiered wet-lab validation staging: Low-risk industrial DNA undergoes miniaturized in vitro testing; high-risk therapeutic or environmental sequences progress through cell culture, small animal models, and contained biocontainment facility trials before scaled deployment.
4. Iterative model safety retraining: All lab failure data (toxicity, unintended function, non-viable constructs) feeds back into Evo’s training loop, gradually reducing the model’s tendency to generate high-risk unplanned sequences over repeated use.

International comparative benchmark: The EU’s Synthetic Biology Regulatory Framework mirrors key elements of Nguyen’s lab protocol, mandating dual-use risk assessments for all AI-generated genetic material, though it lacks standardized genomic AI access tiering seen in Nguyen’s internal research rules.

2D.4 Targeted Solutions and Suggestions (Tiered by Stakeholder Group)

Tier One: Academic & Commercial Synthetic Biology Labs (Immediate, Low-Cost Mitigation)

1. Adopt Evo-style closed-loop DBTL workflows with mandatory in silico risk screening for every de novo DNA candidate prior to synthesis orders.
2. Implement internal tiered AI access permissions: restrict full-capacity Evo model access to senior trained molecular biologists with biosecurity certification; limit student and junior researcher access to pre-filtered low-risk prompt templates only.
3. Standardize comprehensive sequence documentation logs recording all AI prompts, model parameters, risk screening results, and wet-lab validation outcomes for every generated DNA construct.
4. Separate high-risk pathogen-related de novo design work into dedicated BSL-2/3 lab zones with isolated cloud AI hardware disconnected from public internet networks.
5. Retrain internal genomic AI quarterly using local lab failure datasets to reduce generation of unpredictable toxic or non-functional sequences.

Tier Two: Biotech Corporations & DNA Synthesis Vendors (Medium-Term Industry Reform)

1. Integrate real-time API data sharing between genomic AI platforms and synthesis order portals to enable end-to-end automated dual-use screening across digital design and physical manufacturing stages.
2. Establish independent third-party biosecurity audit boards to review high-volume de novo sequence generation workflows quarterly, with public transparency reports on safety compliance.
3. Develop function-based hazard detection algorithms (not just sequence matching) to identify novel pathogenic de novo DNA that evades traditional genetic threat databases.
4. Offer subsidized training programs to build cross-disciplinary teams combining AI engineering, molecular biology, and biosecurity expertise to resolve the field’s critical talent shortage.
5. Limit bulk unvetted de novo synthesis orders to verified academic or pharmaceutical research institutions; block anonymous cloud AI users from ordering large-scale custom DNA batches.

Tier Three: National Regulators & UN Global Biosecurity Bodies (Long-Term Structural Governance)

1. Amend the Biological Weapons Convention to add binding annexes governing access to generative genomic AI and mandatory pre-synthesis function risk screening for all de novo DNA sequences.
2. Create an international shared database of AI-generated hazardous genetic motifs, enabling cross-border real-time screening across all national synthesis vendors and cloud AI providers.
3. Enact uniform national legislation classifying high-capacity genomic foundation models as dual-use biotech tools subject to export control licensing for cross-border model weight sharing.
4. Launch UN-administered low-cost Evo-style model access for low-income research nations paired with free biosecurity training to eliminate global risk inequity from uneven tool access.
5. Mandate public annual safety reporting requirements for all firms operating commercial de novo genomic AI platforms, with financial penalties for incomplete risk screening protocols.

2D.5 Safeguard Measures for Implementation of AI Genomic Governance

1. Phased policy rollout: Introduce digital sequence screening rules over three staggered compliance windows to avoid disrupting ongoing legitimate vaccine, biomanufacturing, and gene therapy research programs.
2. Independent blind biothreat red-teaming: Require annual third-party testing of all genomic AI platforms, where security researchers attempt to elicit hazardous de novo sequences to identify and patch model prompt filter vulnerabilities.
3. Equity funding guardrails: All new global biosecurity regulations must include grant programs for low-resource labs to purchase screening hardware and train biosecurity staff, preventing a two-tier global innovation divide.
4. Permanent cross-stakeholder advisory panels: Form standing committees of synthetic biologists, AI researchers, biosecurity experts, and ethicists to update de novo DNA safety standards every two years as Evo and competing models advance.
5. Immutable sequence logging standards: Mandate encrypted, tamper-proof digital record-keeping for all AI-generated DNA designs to enable full traceability during cross-border biothreat investigations.

Section Three: Application and Implications

3.1 Practical Application Scenarios

Pharmaceutical & Gene Therapy Developers

Teams deploy Evo to design fully novel therapeutic gene circuits, de novo viral delivery vectors, and custom CRISPR editing systems tailored to rare genetic disorders. The closed-loop DBTL workflow slashes traditional multi-year therapy development timelines while layered in silico screening mitigates patient safety risks from untested novel DNA constructs.

Industrial Biomanufacturing Engineers

Scientists generate artificial metabolic gene pathways that enable microbes to produce sustainable biofuels, biodegradable plastics, and pharmaceutical precursors no natural organism can synthesize, drastically cutting reliance on fossil fuel feedstocks. Evo’s long context window designs full multi-gene operons in a single generation step.

Infectious Disease & Public Health Researchers

Laboratories design AI-built synthetic phage genomes that overcome evolving bacterial antibiotic resistance, creating precision anti-pathogen treatments for hospital superbug outbreaks. De novo DNA avoids limitations of natural phage genomes that cannot target rapidly mutating bacterial surface proteins.

Astrobiology & Space Biology Teams

Researchers design extremophile-adapted synthetic microbial DNA sequences for Mars terraforming and off-world life support systems, creating organisms engineered to survive extreme radiation, cold, and low-pressure environments with no Earth evolutionary analogs, a use case Nguyen highlights in his TED talk.

Biosecurity Regulators & Synthesis Vendors

Compliance teams deploy function-based AI screening tools to flag dangerous de novo genetic code before physical DNA production, closing the historic security gap between digital sequence design and chemical synthesis manufacturing.

3.2 Common Misconceptions and How to Avoid Them

1. Misconception: AI-generated DNA sequences will always behave predictably inside living cells. Correction: Nguyen emphasizes all generative genomic outputs require wet-lab validation; statistical pattern matching cannot fully model complex cellular epigenetics, protein folding, and inter-gene interactions that drive emergent unexpected function.
2. Misconception: Traditional sequence-matching biothreat filters can block all hazardous AI-designed DNA. Correction: Evo creates fully novel pathogenic gene cassettes with zero similarity to known threats, rendering homology-based screening useless—only function-prediction algorithms can identify these hidden risks.
3. Misconception: De novo DNA design will quickly enable fully synthetic human or complex animal genomes. Correction: Current Evo architectures only reliably generate microbial-scale sequences; eukaryotic multi-chromosome design remains computationally and experimentally unfeasible in the near term due to unmodeled long-range epigenetic regulation.
4. Misconception: Restricting genomic AI access will stall life-saving medical innovation. Correction: Nguyen’s tiered access framework only limits unsupervised anonymous high-risk design; trained certified researchers retain full access to Evo for vaccine, therapy, and climate biology breakthroughs with standardized safety guardrails.
5. Misconception: Open-source release of full Evo model weights poses minimal biosecurity risk. Correction: Unrestricted public access to complete long-range genomic foundation models lowers the technical barrier for bad actors to generate novel hazardous genetic material without specialized molecular training.

3.3 Key Insights for Readers and Practitioners

Critical Mindset Shifts

All synthetic biology and AI stakeholders must abandon the historic separation between digital sequence design and physical biosecurity risk—every AI-generated DNA string carries tangible real-world biological consequences requiring coordinated oversight. Practitioners must move past viewing genomic AI purely as a speed-up tool and recognize its unique capacity to invent entirely new biological systems with no natural evolutionary precedent, demanding updated ethical and safety frameworks.

Actionable Short-Term Recommendations

• Lab researchers: Integrate two-stage in silico risk screening before submitting any AI de novo sequence for DNA synthesis.
• Biotech leadership: Implement tiered genomic AI access limits and mandatory biosecurity training for all staff working with de novo design tools.
• Regulators: Draft new dual-use rules covering cloud-based genomic AI platforms, not only physical DNA manufacturing facilities.
• Academic educators: Integrate Nguyen’s TED talk and Evo safety protocols into synthetic biology undergraduate curricula to build early biosecurity literacy.

Long-Term Developmental Guidance

Generative genomic AI must be treated as a dual-use strategic technology requiring permanent cross-industry, cross-border safety coordination rather than unregulated open innovation. All future foundation model development should embed pre-generation safety constraints directly into model architecture, rather than relying on post-hoc screening as an afterthought. Global scientific communities must prioritize equitable access to safe genomic AI tools to prevent risk concentration within wealthy industrialized nations alone.

Section Four: Summary and Outlook

4.1 Summary of Core Viewpoints

Eric Nguyen’s TED talk introduces Evo, a breakthrough long-range multimodal genomic AI capable of generating fully original, biologically functional de novo DNA sequences that redefine synthetic biology’s limits, shifting life science from reading natural genetics to intentionally writing custom biological code. Four interconnected risks accompany this innovation: unpredictable emergent function in unvalidated sequences, novel dual-use pathogens undetectable by traditional screening, technical bottlenecks limiting complex eukaryotic genome design, and fragmented global biotech governance creating cross-border safety loopholes rooted in data bias, decoupled design-synthesis regulation, profit-driven speed incentives, and outdated international treaties. Tiered lab, corporate, and UN policy interventions modeled on Nguyen’s closed-loop DBTL pipeline balance transformative medical, climate, and space biology progress with rigorous biosecurity guardrails, requiring mandatory in silico screening, tiered AI access, end-to-end design-synthesis oversight, and updated global dual-use biotech agreements. Long-term safe scaling of de novo genomic AI demands permanent cross-disciplinary safety governance and equitable global tool access to avoid uneven risk distribution worldwide.

4.2 Future Development Trends and Outlook

Emerging Research Directions

Next-generation genomic foundation models will integrate multi-omic epigenetic and structural protein folding data to drastically reduce unforeseen emergent function in de novo DNA, cutting wet-lab validation failure rates further. Interdisciplinary ethics research will build universal global frameworks for ownership and environmental release of AI-designed synthetic organisms, expanding on Nguyen’s TED public discourse. Technical work will extend Evo’s context window to enable partial eukaryotic gene circuit design for mammalian cell therapeutic applications.

New Policy and Industry Challenges

Mass consumer-accessible simplified genomic AI web interfaces will lower technical barriers to de novo DNA design, amplifying dual-use risk for untrained anonymous users without biosecurity training. Rapid expansion of decentralized cloud AI computing makes centralized regulatory monitoring increasingly difficult, requiring automated cross-platform digital screening APIs as a core policy solution. Rising geopolitical competition over synthetic biology supremacy risks fragmented competing national biosecurity standards rather than unified global rules.

Avenues Worth Further Scholarly Exploration

Comparative cross-national analysis of genomic AI regulation will identify the most effective tiered access and screening models for universal UN adoption. Longitudinal lab trial research will quantify long-term environmental safety of AI-designed synthetic microbes for carbon capture and biomanufacturing. Intersectional equity research will measure how uneven global access to Evo-style generative DNA tools widens disparities in pandemic response and climate engineering capacity between high-income and developing nations.

Section Five: References

1. Nguyen, E. (2025). AI that designs entirely new DNA sequences. TED Talk. https://www.ted.com/talks/eric_nguyen_ai_that_designs_entirely_new_dna_sequences
2. Nguyen, E. T. et al. (2024). Sequence modeling and design from molecular to genome scale with Evo. Science. https://www.science.org/doi/10.1126/science.ado9336
3. Hazy Research Lab. (2024). OpenGenome Dataset & Evo Model Technical Whitepaper. https://hazyresearch.stanford.edu/blog/2024-03-14-evo
4. National Academies of Sciences. (2026). Dual-Use Risks of Generative Genomic Artificial Intelligence. NCBI Accession NBK614591.
5. European Chemicals Agency. (2025). Synthetic Biology & AI Sequence Oversight Regulatory Guidelines.
6. TED Talks Daily Podcast. (2025). Interview with Eric Nguyen on de novo DNA AI innovation and biosecurity risks. https://wave.co/podcast/ted-talks-daily/how-ai-could-generate-new-life-forms-eric-nguyen-30e3e3da
7. Cell Systems Journal. (2026). Limitations of Current Generative Genomic Models for Full Genome Design.
8. Biological Weapons Convention UN Secretariat. (2025). Working Paper on AI-Driven Synthetic DNA Biothreats.

Short Encouraging Exploration Note

Review Eric Nguyen’s open-source Evo research documentation to examine real lab-generated novel DNA sequences, and watch his full TED talk to deepen understanding of the ethical tradeoffs between unrestricted genomic AI innovation and global biothreat protection.