Synthesis screening evasion
AI-redesigned sequences keep their function while no longer resembling what synthetic DNA providers screen for.
- Severity
- Catastrophic
- Horizon
- 1–3 years
- Evidence
- Lab
- Consensus
- Medium
Ordering synthetic DNA from a commercial provider is the only moment in the process at which a third party can look without knowing what it is for. That control works by comparing the requested sequence against lists of agents of concern. The risk is that AI-assisted design produces sequences that keep the function and stop resembling the reference.
It is demonstrated in the lab, and it has a name. Wittmann and eleven co-authors published in Science, on 2 October 2025, an evaluation of open-source AI-enhanced protein design software: they identified a vulnerability by which redesigned sequences were not reliably detected, and developed and deployed patches [1114]Strengthening nucleic acid biosecurity screening against generative protein design toolsView source ↗. In 2026 RAND tested autonomous agentsAI agentAn AI system that does more than answer: it takes a goal and acts on its own to reach it, step by step, using tools such as a browser, email or a terminal, without anyone approving each step.For exampleAsking an assistant to suggest flights is using a chatbot. Asking it to search, compare, buy the ticket and put it in your calendar, all by itself, is using an agent. operating biological tools: some redesigned sequences evaded screening, with low reliability —Gemini 3.1 Pro managed it in 3 of 30 attempts— [888]Testing Large Language Model Agents on the Use of Biological Tools for Nucleic Acid Synthesis Screening EvasionView source ↗.
The regulatory framework does not keep up. The OSTP one dates from April 2024 and operates as a funding condition, not as law [833]Framework for Nucleic Acid Synthesis ScreeningView source ↗archived copy only; the executive order of May 2025 ordered it replaced within 90 days [364]Improving the Safety and Security of Biological ResearchView source ↗ and thirteen months past the deadline the official page still announces that it will be updated [101]Synthetic Nucleic Acid ScreeningView source ↗.
What this does not demonstrate. No documented evasion ended in manufactured material: the two known cases are evaluation exercises whose findings were patched. RAND’s finding 5 points in an uncomfortable direction for both camps: the providers’ filters made it impossible to test the most advanced closed models, so nobody —not even the external evaluators— knows what the real ceiling is on this task [888]Testing Large Language Model Agents on the Use of Biological Tools for Nucleic Acid Synthesis Screening EvasionView source ↗. And Anthropic announces that it will deprecate its own screening-evasion evaluation because it doubts that its results are a meaningful proxy for biological viability [82]System Card: Claude Fable 5.1 & Claude Mythos 5.1View source ↗.
Chain of materialisation
PreconditionObserved
Screening is voluntary, partial and unaudited
The 2024 OSTP framework operates as a federal funding condition, not law, and the May 2025 executive order required its revision within 90 days. Thirteen months past the deadline, the ASPR page still says it will be updated once the revised framework is available.
Precedents: Open letter for mandatory nucleic acid synthesis screening
TriggerLab
AI redesign breaks sequence matching
Wittmann and co-authors evaluated open-source AI protein design software in Science and found that AI-redesigned sequences were not reliably detected by current screening tools. Patches were developed and deployed; the cycle was handled as a responsible-disclosure exercise.
Precedents: A synthesis-screening evasion found via AI protein design is patched · The Fable 5.1 and Mythos 5.1 system card reports control circumvention in production · The GPT-5.6 system card documents misalignment in internal deployment
Observed and demonstrated evidence ends here. What follows is projection.
CascadeProjected
An evasive order reaches a provider that does not screen
The International Gene Synthesis Consortium claims to represent most global commercial capacity, but benchtop synthesis is outside the perimeter and no one audits compliance. That a redesigned order crosses that gap is plausible and undocumented.
ImpactSpeculative
Material of concern manufactured with no alarm raised
Screening is the only point in the process where a third party can intervene without knowing the customer's intent. If it stops seeing, the rest of the chain is private. Harm magnitude depends on the agent and there is no case to extrapolate from.
Scenarios where it appears
Related measures
See on the map →Report a mistake in this entry →
Sources
- [1114] Strengthening nucleic acid biosecurity screening against generative protein design tools · Microsoft Research 2025
- [723] Strengthening biosecurity in the era of AI · Microsoft 2026
- [888] Testing Large Language Model Agents on the Use of Biological Tools for Nucleic Acid Synthesis Screening Evasion · RAND Corporation 2026
- [833] Framework for Nucleic Acid Synthesis Screening · The White House (OSTP) 2024 archived copy only
- [364] Improving the Safety and Security of Biological Research · The White House 2025
- [101] Synthetic Nucleic Acid Screening · HHS / ASPR 2026
- [547] International Gene Synthesis Consortium · International Gene Synthesis Consortium 2026
- [944] An Open Letter in Support of Mandatory Nucleic Acid Synthesis Screening and Recordkeeping · Coalición de firmantes de tecnología, industria de síntesis, ciencias de la vida y seguridad nacional 2026
- [82] System Card: Claude Fable 5.1 & Claude Mythos 5.1 · Anthropic 2026
- [825] GPT-5.6 System Card · OpenAI 2026