March 8, 2024

Advances in artificial intelligence (AI) are creating unprecedented opportunities for life science research, including by enabling the design of functional biological molecules, especially proteins. This application of AI for protein design holds immense potential to enhance our understanding of the world and help address some of humanity’s most pressing challenges by enabling rapid responses to infectious disease outbreaks, curing numerous diseases, unlocking sustainable sources of energy, helping to mitigate climate change, and more.

As scientists engaged in this work, we believe the benefits of current AI technologies for protein design far outweigh the potential for harm and we would like to ensure our research remains beneficial for all going forward. Given anticipated advances in this field, a new proactive risk management approach may be required to mitigate the potential of developing AI technologies that could be misused, intentionally or otherwise, to cause harm. We are therefore motivated as a community to articulate a set of values and principles to guide the responsible development of AI technologies in the field of protein design. These values include safety, security, equity, international collaboration, openness, responsibility, and pursuing research for the benefit of society. Furthermore, we as signatories voluntarily agree to a set of specific, actionable commitments informed by these values and principles and outlined here. We will work together with global stakeholders across academia, governments, civil society, and the private sector to ensure that this technology develops in a responsible and trustworthy manner and that it is safe, secure, and beneficial for all.

Values and Guiding Principles

For the benefit of society — The responsible development of AI technologies for protein design has the potential to generate new knowledge and help solve some of the world’s most critical problems. Our community will work with governments, civil society, funders of science, academics, and the private sector to ensure that our research benefits society, including by using protein design to help defend against pandemics and other relevant emergencies. We will pursue applications that seek to promote health, sustainability, security, and scientific knowledge and refrain from those likely to result in harm, whether accidental or intentional. To build and maintain trust, we will engage stakeholders with diverse perspectives and inform the public of the benefits, limitations, risks, and risk mitigation strategies of our work.

Safety and Security — Since no computationally designed protein can cause real-world harm unless it is physically produced, the manufacturing of synthetic DNA presents a key biosecurity checkpoint for the field of computational protein design. We will participate in and endeavor to improve the process of nucleic acid synthesis screening, with the aim of detecting hazardous biomolecules before they can be manufactured. In addition, providing access to software is essential to advance science, but we as developers have a duty to assess and mitigate risks that could emerge from these systems before they are widely shared. We believe the benefits of openly sharing scientific methods strongly outweigh potential risks; nevertheless, our community will continuously assess our protein design software, including how these tools intersect with other technologies, and apply appropriate safety and security measures. Such assessments should be grounded in evaluations by scientific experts with relevant domain knowledge and seek to determine the extent to which the release of a given AI model, its model weights, or the datasets or methods used to train the model would introduce identifiable risks in the context of existing risks and risk mitigation strategies. All efforts to evaluate security, including red teaming, should be undertaken with precaution to avoid creating roadmaps for misuse. Safety and security risks should never be overstated or used as an unjustified excuse to limit open science.

Openness — Many researchers in our community benefit from open-source scientific software, which has enabled rapid innovation and broad collaboration. We affirm the core scientific value of openness and believe clear standards around openness will help AI technologies develop along a path that maximizes benefits to society. Decisions around the release of AI systems for protein science should start with the guiding principle of scientific openness and be informed by efforts to identify and mitigate meaningful risks prior to release. Practices for limiting access and distribution should be followed for AI systems that present identified meaningful and unresolved risks. Such practices should be consistent with the principles outlined here, and such risks should be disclosed to appropriate audiences.

Equity and International Collaboration — For the field of protein design to deliver the greatest benefit to society, it should have broad stakeholder input and diverse participation in research directions. We will work toward greater inclusion of the global community to enable more equitable participation in the research itself and its benefits. The protein design community is international, and the benefits of our research for the world are greater when conducted as an international effort. As scientists, we affirm our membership in a global community rooted in shared values, including adherence to the scientific method. We will work collaboratively with stakeholders from around the world to advance the responsible development of AI for the world and, as individuals, adhere to commitments on responsible AI development enacted by our respective governments.

Responsibility — The path that technology takes depends on what we as researchers choose to work on. While recognizing that most advanced technologies can be harnessed for good or deliberately misused to do harm, we will refrain from research that is likely to lead to overall harm or enable misuse of our technologies. We will work together to establish responsibility as a core tenet of community practice and develop and share best practices for preventing misuse of our technologies. Our community will participate in developing the detailed processes needed to carry out our commitments. Moreover, we will continuously reevaluate how these values, guiding principles, commitments, and policies are implemented by our community and others to ensure responsiveness to new developments.

Commitments to Drive Responsible AI Development

We as signatories agree to the following voluntary commitments. Each commitment is accompanied by a list of possible implementation strategies, though these lists are not exhaustive. Each signatory should enact these commitments in ways that are appropriate for them.

1. We will conduct research for the benefit of society and refrain from research that is likely to cause overall harm or enable misuse of our technologies.

This can be implemented by pursuing research that seeks to generate new knowledge, promote health and well-being, achieve sustainability, justice, or equity, or otherwise advance human progress; and by working with governments, civil society, funders, and other stakeholders to ensure that our research is aligned with these goals.

2. We will support community efforts to prepare for and respond to infectious disease outbreaks and other relevant emergencies.

This can be implemented by organizing, participating in, or otherwise supporting response teams for coordinated scientific action, such as rapid countermeasure development in the event of an infectious disease outbreak; by conducting research into priority pathogens; by working to shorten the time needed to create safe and effective countermeasures, including diagnostics, medicines, and vaccines; or by otherwise supporting these efforts.

3. We will obtain DNA synthesis services only from providers that demonstrate adherence to industry-standard biosecurity screening practices, which seek to detect hazardous biomolecules before they can be manufactured.

This can be implemented by procuring synthetic DNA from manufacturers that perform appropriate safety screening; by creating, sharing, and adhering to a list of such manufacturers; by requiring such screening as a condition for publication; or by supporting policies that require such screening.

4. We will support the development of new strategies to improve DNA synthesis screening, with the aim of better detecting hazardous biomolecules before they can be manufactured.

This can be implemented by actively collaborating with DNA synthesis stakeholders where we have relevant expertise, including by creating AI tools for the improved detection of potentially hazardous biomolecules; by participating in secure evaluations and red team exercises that seek to identify gaps in current screening protocols; or by otherwise supporting these efforts.

5. We will continuously evaluate the capabilities of our protein design software and seek to identify and mitigate all meaningful safety and security risks.

This can be implemented by conducting evaluations of AI systems to identify meaningful safety and security concerns prior to release; by resolving identified meaningful risks before release; and by limiting the release of AI systems with identified meaningful risks that cannot be resolved.

6. We will support efforts to improve the methods by which our protein design software is evaluated to better identify risks.

This can be implemented by participating in research to improve the theoretical understanding of risks; by engaging in joint development of evaluation frameworks; by critically assessing the performance of current evaluation frameworks; engaging with experts in other relevant domains; or by otherwise supporting these efforts.

7. We will review the capabilities and risks in our field at regular, secure meetings.

This can be implemented by organizing and participating in gatherings that include scientists, representatives from regulatory bodies, and other stakeholders to review developments in the field, including results from various risk assessments, to assess current and future concerns.

8. We will report concerning research practices.

This can be implemented by always adhering to ethical research practices; by using appropriate systems to report concerning behavior, such as the intent to use protein design for harm; or by reporting on concerning research practices in the field to appropriate authorities, including any failure to abide by relevant laws.

9. We will communicate the benefits and risks of our research.

This can be implemented by including appropriate disclosures of relevant benefits, risks, and risk mitigation strategies when sharing or publishing new software; or by providing updates to and responding to requests from regulatory bodies and other stakeholders as needed.

10. We will revise these principles and commitments as needed and share them with new members in the field.

This can be implemented by maintaining a culture of responsibility; by reviewing these guidelines and best practices with colleagues, including students and trainees; by incorporating discussion of responsible behavior in scientific conferences; and by participating in community-wide efforts to evaluate risks and assess current guidelines and commitments as our field continues to advance.

Signatories

Affiliations are for identification only and do not imply any institutional endorsement.

• Carlos G. Acevedo-Rocha, DTU Biosustain, Denmark
• Amirali Aghazadeh, Georgia Tech, USA
• Mohammed AlQuraishi, Columbia University, USA
• Juan Alvarez, Merck Research Laboratories, USA
• Ingemar André, Lund University, Sweden
• Frances Arnold — 2018 Nobel Laureate, California Institute of Technology, USA
• Aaron Arvey, Third Rock Ventures, USA
• Mihai L. Azoitei, Duke University, USA
• Minkyung Baek, Seoul National University, South Korea
• David Baker, University of Washington, USA
• Patrick Barth, Swiss Federal Institute of Technology Lausanne & Baylor College of Medicine, Switzerland & USA
• Regina Barzilay, Massachusetts Institute of Technology, USA
• Ariel J. Ben-Sasson, ZipBio, Israel & USA
• Gaurav Bhardwaj, University of Washington, USA
• Thomas Bourquard, MAbSilico, France
• Michael Bronstein, University of Oxford, UK
• Benjamin Brown, Vanderbilt University, USA
• Jennifer Bui, University of British Columbia, Canada
• Adrian Bunzel, Federal Institute of Technology Zurich, Switzerland
• Chris Bystroff, Rensselaer Polytechnic Institute, USA
• Alessandra Carbone, Sorbonne University, France
• Pranam Chatterjee, Duke University, USA
• Zibo Chen, Westlake University, China
• Jianlin Cheng, University of Missouri, USA
• George Church, Harvard University, USA
• Bruno Correia, Swiss Federal Institute of Technology Lausanne, Switzerland
• Juan Cortés, University of Toulouse, France
• René M. de Jong, dsm-firmenich, Netherlands
• Cesar de la Fuente-Nunez, University of Pennsylvania, USA
• Erika DeBenedictis, Francis Crick Institute, UK
• William Degrado, University of California, San Francisco, USA
• Adji Bousso Dieng, Princeton University, USA
• Hendrik Dietz, Technical University of Munich, Germany
• Sijia Dong, Northeastern University, USA
• Jiayi Dou, ShanghaiTech University, China
• Kevin Drew, University of Illinois Chicago, USA
• Frédéric Dreyer, Exscientia AI, UK
• Joshua Dunn, Ginkgo Bioworks, USA
• Arne Elofsson, Stockholm University, Sweden
• Tim Ermak, Silica Corpora, Germany
• Daniel Alejandro Fernandez Velasco, National Autonomous University of Mexico, Mexico
• Andrew Ferguson, University of Chicago, USA
• Noelia Ferruz, Barcelona Institute of Molecular Biology, Spain
• Eric Fischer, Harvard University, USA
• Polly Fordyce, Stanford University, USA
• Mark Gerstein, Yale University, USA
• Molly Gibson, Flagship Pioneering & Generate Biomedicines, USA
• Jimmy D. Gollihar, Houston Methodist Research Institute, USA
• Glen Gowers, Basecamp Research, UK
• Jeffrey Gray, Johns Hopkins University, USA
• Gevorg Grigoryan, Generate Biomedicines, Dartmouth College, USA
• Dominik Gront, University of Warsaw, Poland
• Haotian Guo, Ailurus Bio, UK
• Alican Gulsevin, Butler University, USA
• Ankit Gupta, Ginkgo Bioworks, USA
• Shozeb Haider, University College London, UK
• Joseph Harrison, University of the Pacific, USA
• Brian Hie, Stanford University, USA
• Eric Horvitz, Microsoft, USA
• Parisa Hosseinzadeh, University of Oregon, USA
• Jie Hou, Saint Louis University, USA
• Patrick Hsu, University of California, Berkeley & Arc Institute, USA
• Possu Huang, Stanford University, USA
• Thomas Huber, Australian National University, Australia
• Christopher Ing, ProteinQure, Canada
• John Ingraham, Generate Biomedicines, USA
• Tommi Jakkola, Massachusetts Institute of Technology, USA
• Ramesh Jha, Los Alamos National Laboratory, USA
• Shuiwang Ji, Texas A&M University, USA
• Lin Jiang, University of California, Los Angeles, USA
• Joe Kaczmarski, Australian National University, Australia
• John Karanicolas, Fox Chase Cancer Center, USA
• Panagiotis Kastritis, University of Halle-Wittenberg, Germany
• Amy Keating, Massachusetts Institute of Technology, USA
• Sagar Khare, Rutgers University, USA
• Firas Khatib, University of Massachusetts Dartmouth, USA
• Alena Khmelinskaia, Ludwig Maximilian University of Munich, Germany
• Doo Nam Kim, Pacific Northwest National Laboratory, USA
• Ho Min Kim, Korea Advanced Institute of Science and Technology, South Korea
• Philip M. Kim, University of Toronto, Canada
• Neil King, University of Washington, USA
• Ross D. King, Chalmers University, Sweden
• Nobuyasu Koga, Osaka University, Japan
• Maksym Korablyov, DreamFold & Quebec AI Institute, Canada
• Tanja Kortemme, University of California, San Francisco, USA
• Brian Kuhlman, University of North Carolina at Chapel Hill, USA
• Daniel Kulp, Wistar Institute, USA
• Daisuke Kuroda, National Institute of Infectious Diseases, Japan
• Elodie Laine, Sorbonne University, France
• Henry Lee, Cultivarium, USA
• Bruce Lichtenstein, University of Portsmouth, UK
• Zeming Lin, New York University, USA
• Steffen Lindert, Ohio State University, USA
• Chenghao Liu, DreamFold & Quebec AI Institute, Canada
• Sen Liu, Hubei University of Technology, China
• Yi Liu, Stony Brook University, USA
• Ajasja Ljubetič, National Institute of Chemistry , Slovenia
• Peilong Lu, Westlake University, China
• Jeremy Luban, University of Massacusetts, USA
• Ali Madani, Profluent Bio, USA
• Tobias Madl, Medical University of Graz, Austria
• Enrique Marcos, Spanish National Research Council, Spain
• Debora Marks, Harvard University, USA
• Jens Meiler, Leipzig University & Vanderbilt University, Germany & USA
• Hans Melo, Menten AI, USA
• James Moody, Brigham Young University, USA
• Grant S. Murphy, Merck Research Laboratories, USA
• David Nannemann, Rosetta Design Group, LLC, USA
• Hafumi Nishi, Tohoku University, Japan
• Masha Niv, Hebrew University of Jerusalem, Israel
• Pascal Notin, Harvard University, USA
• Gustav Oberdorfer, Graz University of Technology, Austria
• Sílvia Osuna, University of Girona & Catalan Institution for Research and Advanced Studies, Spain
• Carlos Outeiral, Oxford University, UK
• Sergey Ovchinnikov, Massachusetts Institute of Technology, USA
• Hahnbeom Park, Korea Institute of Science and Technology, South Korea
• John Parkhill, Terray Therapeutics, USA
• Fabio Parmeggiani, University of Bristol, UK
• Brian Pierce, University of Maryland, USA
• Douglas Pires, University of Melbourne, Australia
• Fabien Plisson, Center for Research and Advanced Studies of the National Polytechnic Institute, Mexico
• Francesco Raimondi, Scuola Normale Superiore, Italy
• Arvind Ramanathan, Argonne National Laboratory & University of Chicago, USA
• Cesar Ramírez-Sarmiento, Pontifical Catholic University of Chile, Chile
• Rama Ranganathan, University of Chicago, USA
• Alexander Rives, New York University, USA
• Sergio Romero-Romero, University of Bayreuth, Germany
• Jeff Ruffolo, Profluent Bio, USA
• Andrew Satz, EVQLV, USA
• William Schief, Moderna Therapeutics, Scripps Research Institute, & IAVI, USA
• Torben Schiffner, Scripps Research Institute, USA
• Clara Schoeder, Leipzig University, German
• Ora Schueler-Furman, Hebrew University, Israel
• Ingmar Schuster, Exazyme, Germany
• Petra Schwille, Max Planck Institute of Biochemistry, Germany
• Fatemeh Shirazi, Microvi Biotech Inc., USA
• Raghav Shroff, Houston Methodist Research Institute, USA
• Justin B. Siegel, University of California, Davis, USA
• Geoffrey H. Siwo, University of Michigan, USA
• Joanna Slusky, University of Kansas, USA
• Colin A. Smith, Wesleyan University, USA
• Vanita D. Sood, Fable Therapeutics, Canada
• Amelie Stein, University of Copenhagen, Denmark
• Lance Stewart, University of Washington, USA & Denmark
• Eva-Maria Strauch, Washington University, USA
• Alexander Titus, In Vivo Group & University of Southern California, USA
• Tom Thompson, University of Cincinnati, USA
• Wouter Touw, dsm-firmenich, Netherlands
• Stef van Grieken, Cradle, Switzerland & Netherlands
• Darya Vanichkina, University of Sydney, Australia
• Anastassia Vorobieva, Flemish Institute for Biotechnology & Free University of Brussels, Belgium
• Anna Marie Wagner, Ginkgo Bioworks, USA
• Newton Wahome, Coalition for Epidemic Preparedness Innovations, USA
• Chu Wang, Peking University, China
• Kathy Wei, 310 Inc, USA
• Martin Weigt, Sorbonne University, France
• Andrew White, University of Rochester, USA
• Bruce Wittmann, Microsoft, USA
• Christopher W. Woods, University of Edinburgh, UK
• Emily Wrenbeck, Ginkgo Bioworks, USA
• Dong Xu, University of Missouri, USA
• Zhongyue (John) Yang, Vanderbilt University, USA
• Vladimir Yarov-Yarovoy, University of California, Davis, USA
• Yang Zhang, National University of Singapore, Singapore

About

This community statement originated from discussions that occurred during and after a summit at the University of Washington Institute for Protein Design on October 25, 2023. Those conversations included leading academics from around the world and representatives from several private sector, philanthropy, and government organizations.

The statement was prepared by Sarah R. Carter‌, Samuel Curtis, Claudia Emerson, Jeffrey Gray, Ian C. Haydon, Andrew Hebbeler, Claire Qureshi, Nicholas Randolph, Alexander Rives, and Lynda Stuart.