AI Executives and Biosecurity Experts Urge Mandatory Screening for Gene Synthesis Orders

AI company leaders including Sam Altman of OpenAI, Demis Hassabis of Google DeepMind, and Dario Amodei of Anthropic have differed on multiple issues, including the pace of AI development, regulatory approaches, and societal preparation for advanced AI systems. Despite these differences, the three executives joined 85 other specialists in technology, biology, and national security in signing an open letter that recommends stronger requirements for screening gene synthesis orders. The letter focuses on risks that AI tools could assist in designing or producing biological agents through the chemical assembly of custom DNA sequences.

Gene synthesis enables laboratories to produce specified DNA strands without relying on natural templates. Commercial providers supply these sequences for vaccine development, therapeutic proteins such as insulin, gene therapies for conditions including hemophilia, agricultural applications, CAR-T cancer treatments, and diagnostic reagents. Demand for synthetic DNA has increased as costs have declined.

The same process that supports these applications can also be used to assemble genetic material associated with pathogens. Most United States gene synthesis companies currently screen customer orders against lists of sequences linked to heightened pathogenicity or transmissibility and verify customer identities. These practices remain voluntary. Becky Mackelprang, director for security programs at the Engineering Biology Research Consortium, stated that voluntary screening leaves open the possibility that some providers will not implement checks.

James Diggans, vice president of policy and biosecurity at Twist Bioscience, noted that commercial gene synthesis has operated for more than twenty years without documented misuse resulting in harm. AI systems introduce additional considerations because large language models and specialized biodesign tools can generate novel genetic sequences that differ from known pathogens. Existing screening methods rely on sequence similarity to flagged entries; novel sequences may therefore evade detection. A 2025 study in Science found that current screening methods have matched AI capabilities to date, though industry representatives have indicated that this alignment may not persist indefinitely.

Mackelprang observed that AI systems have performed well on queries about laboratory procedures that would normally require specialized training. She added that practical laboratory execution still demands hands-on experience and that AI assistance would not enable individuals lacking biological training to produce significant hazards. The primary remaining control point remains the gene synthesis provider. Mackelprang’s principal concern is that AI could be used to create sequences intended to bypass screening protocols. She assessed the near-term probability of such misuse as low but stated that the potential consequences warrant development of preventive measures.

The open letter, organized by the Institute for Progress and the Foundation for American Innovation, calls for mandatory screening and retention of order records. Signatories include gene synthesis companies, technology executives, life-science industry representatives, and national security specialists. The letter recommends that records be maintained to permit tracing of any sequence that evades initial screening. It also endorses the bipartisan Biosecurity Modernization and Innovation Act, which would direct the Department of Commerce to establish screening standards within one year, and suggests that states adopt consistent requirements aligned with federal and industry guidelines.

The letter does not propose applying biosecurity rules directly to AI developers. Major AI companies already implement measures to limit model outputs that could provide detailed assistance with biological agents, though these safeguards are not uniformly effective. The letter instead emphasizes screening at the point of DNA synthesis as a concrete step supported by multiple providers.

Anthropic has posted a position for a technical investigator focused on chemical, biological, radiological, and nuclear threats. OpenAI launched GPT-Rosalind for life-sciences research and introduced a related biodefense program for trusted developers. On June 4, 2026, personnel from both companies participated in a Bipartisan Commission on Biodefense session addressing AI and biological risks.

Twist Biosciences’ Diggans stated that AI systems can also serve defensive functions by evaluating orders for engineered sequences with the same scrutiny applied to natural sequences. David Haussler of the UC Santa Cruz Genomics Institute, another signatory, said that providers should subject orders to review by AI models capable of identifying attempts to construct dangerous pathogens.

Several defensive AI projects are already active. OpenAI provided $30 million in seed funding to Valthos, a startup developing AI systems for threat detection and countermeasure design; Valthos co-founder Kathleen McMahon signed the open letter. In September 2025, the Coalition for Epidemic Preparedness Innovations and Sentinel Bio launched the Pandemic Preparedness Engine, which incorporates biosecurity considerations during system design, including controls on training data, user access, and prompt monitoring. The platform uses AI agents to assess prompts for potential misuse or attempts to circumvent restrictions.

Anthropic released Claude Fable 5, a version of its model configured for broader public access. The system redirects queries involving cybersecurity, biology, chemistry, or model distillation to lower-capability versions. Some users have reported that legitimate biology-related queries are also redirected. Haussler noted that overly broad restrictions can limit beneficial applications while still leaving open the possibility that a successfully jailbroken model could disseminate detailed instructions for constructing dangerous agents.

The letter’s signatories argue that mandatory screening combined with recordkeeping and AI-assisted evaluation at synthesis providers offers a practical method to address risks arising from AI-assisted sequence design. The recommendations focus on requirements applicable to DNA providers rather than on new obligations for AI companies themselves.