| Confronting The Threat Of Ethnic Bioweapons thelastamericanvagabondcom | MR Online Confronting The Threat Of Ethnic Bioweapons. (Photo: thelastamericanvagabond.com)

Dual-use dilemma: AI and the future of protein design

Originally published: Peoples Democracy on April 7, 2024 by S Krishnaswamy (more by Peoples Democracy)  | (Posted Apr 09, 2024)

GIVEN current artificial intelligence (AI) generative and transformative tools, the potential to design proteins in a quick and easy way has increased the possibility of using such synthetic proteins as bio weapons. Concerned about this scenario, a group of scientists initiated a campaign advocating for the responsible and ethical application of protein design.

In October 2023, David Baker of the Institute for Protein Design at the University of Washington, USA, organised an AI safety summit to assess the risks associated with the malicious utilisation of designer proteins. The focus was on determining the necessity and scope of regulation for protein design and identifying potential hazards.

On March 8, 2024, the collective declaration by over 100 researchers, among them a Nobel laureate, was announced in Boston, USA during a conference dedicated to bio molecular engineering. As on March 21, 2024 when the appeal for endorsement was closed, there are 158 signatories and 41 supporters. Most are from USA and Europe. There are a few from China, Korea, Israel and Africa. No organisation is represented. The declaration urges the scientific community to adhere to specific safety and security protocols when employing artificial intelligence in the creation of synthetic proteins.


Proteins are long chains made of amino acid units. These are the workhorses of all viruses, bacteria, plants and animals. They can be thought of as nano machines that carry out their specific functions in precise and efficient ways. When these are mutated or modified, the organism dies or develops problems or gets modified. Over the billions of years, proteins have evolved and have been used as weapons, energy generators, promoters, transporters and so on for all the different types of interactions that go on in and amongst biological entities. For example, it has been shown that a venom toxin—a protein—used by a spider and a centipede, has been re-purposed from an ancestral protein which was functioning as a hormone as seen in a prawn also belonging to the arthropod family.

There are 20 different amino acids and these are like the letters of the alphabet. Just like with our language, the letters of the alphabet can be used to form different sentences. Depending on the sequence of amino acid units, proteins take a specific shape and this helps the protein carry out its function. Till about five years ago the problem of predicting the structure, given a sequence of the amino acids in the protein, was not so easy. But this changed with the coming of a type of AI based deep learning programme called Alpha Fold which was remarkably successful in coming up  with protein models, for a large number of proteins, that were correct in comparison to their experimentally determined protein structure. But like most AI learning tools, it was limited to what was already known.

However, a couple of years ago, it became possible to design quite successfully functional but completely novel or synthetic proteins as researchers developed protein design tools such as RFdiffussion which used AI generative capabilities of diffusion networks. These are based on the same principles as the neural networks that generate realistic images, for example, in the graphics program DALL·E. Training on data such as images or protein structures, these ‘diffusion’ networks undergo a process where the data is gradually distorted until it no longer resembles the original image or structure. Subsequently, the network is trained to ‘denoise’ the data by reversing this process.

Using DNA synthesising technologies people were able to make the DNA corresponding to designed proteins. They then showed in the lab, by putting the DNA into cells to express and analyse the protein, the function of that protein was as designed. So in principle it became possible to have functional synthetic proteins on demand.

The latest protein design tools have shown impressive effectiveness in creating proteins that can carry out precise functions, especially those dictated by a specific shape like binding to a protein surface. Nevertheless, current tools are still unable to meet a wide range of needs, such as developing a protein that can perform a specific reaction regardless of its shape. This constraint becomes evident when the desired function is identified, but the exact geometry is uncertain.


The end of October 2023, the White House in USA issued an Executive Order concerning the Safe, Secure, and Reliable Development and Utilisation of Artificial Intelligence. President Biden of USA emphasized the importance of managing the significant risks associated with leveraging AI for positive purposes and highlighted the necessity for a collaborative effort across society involving government, private industry, academia, and civil society. The accompanying fact sheet outlined the White House’s commitment to establishing new guidelines for AI safety and security to prevent the misuse of AI in creating hazardous biological substances through the implementation of robust screening standards for biological synthesis. A time line of implementation of measures was put forward.

The community of scientists who made the declaration as researchers actively involved in this domain, are confident in the substantial advantages offered by current AI technologies in protein design, far surpass any potential risks. They feel that with the anticipated progress in this field, there is a growing need for a proactive risk management strategy to pre-empt the emergence of AI technologies that could be exploited, either intentionally or inadvertently, for harmful purposes. As a collective, they wish to define a set of core values and principles to steer the ethical advancement of AI technologies within protein design. These values encompass safety, security, fairness, global cooperation, transparency, accountability, and the pursuit of research for societal welfare. They also willingly commit to a series of tangible actions with stakeholders worldwide from academia, government, civil society, and the private sector. They have agreed to foster the open, equitable, responsible and trustworthy evolution of this technology, ensuring its safety, security, and benefit for all. They have agreed to conduct research for the benefit of society and refrain from research that is likely to cause overall harm or enable misuse of our technologies.

Any designed protein to be tested for function and put to real world use needs to have the corresponding DNA synthesised. As on now this capability to do so lies with only a limited set of players. So the researchers have pledged to obtain DNA synthesis services only from providers that demonstrate adherence to industry-standard bio security screening practices, which seek to detect hazardous bio molecules before they can be manufactured. They have agreed to support the development of new strategies to improve DNA synthesis screening, with the aim of better detecting hazardous bio molecules before they can be manufactured.


The process behind the list of principles and voluntary commitments parallels what came out of the Asilomar conference on February 27, 1975 where about 150 molecular biologists then drew up guidelines for the safe use of recombinant DNA. About 90 of the scientists were American; another 60 came from 12 different countries mostly from Europe and Japan, USSR and Israel. No organisations or governments were represented per se. The Asilomar declaration gave rise to many debates and protests especially in the U.S. But it could do little to alter the course of how genetic engineering and recombinant DNA grew into multinational corporations and impacted various aspects of medical health and agriculture across the world. It could not anticipate the PCR and much later the CRISPR revolution that became game changers in the way genetic modification of organisms could be done.

The declarations fail to recognise the larger issues that control and drive science and technologies—the socio-economic-political system in conjunction with the unpredictable development of basic science and its application. Like with atomic power as a weapon of mass destruction—when there is political and economic push, scientists after all are human.

As Prabir Purkayastha says in his book Knowledge as Commons: Towards inclusive science and technology:

Today, the need for organising scientists to struggle for a more democratic scientific decision-making process must go hand-in-hand with a strong movement to bring science to the people. If global warming is to be combated, or nuclear disarmament pursued, it is not enough for scientists to say so. Science has to be brought out of the ivory tower and de-mystified so that the people affected by such decisions can assert their voice. Science is too serious a business to be left to the scientists—it must be a part of our larger struggle for equity and democracy in society.

Monthly Review does not necessarily adhere to all of the views conveyed in articles republished at MR Online. Our goal is to share a variety of left perspectives that we think our readers will find interesting or useful. —Eds.