Researchers taught AI to create novel viruses. What’s next?

What happened: Researchers from Stanford University and the Arc Institute recently taught an AI model to create novel viruses. They trained the AI on genetic sequencing data from millions of plants, animals, viruses, and other living organisms until it was finally able to use that knowledge to write new genetic code.

Why it matters: While the study was limited to viruses that infect only bacteria, the framework could eventually be expanded to include plants, animals, or even humans. If (or when) the technology reaches the point where that is possible, what can be done to prevent it from being used to kill rather than cure?

The backstory: AI created a virus

Scientists at Stanford University and the Arc Institute recently trained their Evo artificial intelligence model to recognize patterns in naturally forming DNA structures and use that information to create new viruses. As Carl Zimmer describes, the new development raises “hopes for medical advances while also raising the disturbing possibility that the technology could someday be used to invent dangerous pathogens.”

In a moment, we’ll discuss how they did it, how real a threat these new developments actually pose, and what—if anything—the government or researchers can do to safeguard these developments. But first, take a moment to reflect on how that news makes you feel. What was your initial reaction to the headline, as well as the basics of what those researchers have accomplished? 

Keep your answer in mind, because I want us to revisit that thought in a minute. For now, though, let’s take a closer look at the AI-engineered viruses and their possible implications.

How good is the AI?

The scientists behind these developments trained their AI system in a similar way to how ChatGPT and other large language models learn. As Zimmer writes, 

DNA is strikingly similar to a book in some ways: a string of molecular building blocks, known as nucleotides, arrayed like letters in a line of text. A gene consists of hundreds of nucleotides drawn from a four-letter alphabet: A, C, G and T. The sequence encodes the instructions for building proteins and other molecules.

Where Evo shone in ways that non-AI attempts at decoding that book have failed—or at least stalled—is in its ability to decode the rules that make sense out of all those letters. By training on genetic sequences from millions of animals, plants, viruses, and other organisms, the AI eventually learned those rules well enough to start generating its own unique genetic codes.

Now, it’s important to note that the scientists behind these developments had the foresight to constrain the AI to viruses that target bacteria, at least for now. Part of the reason is that bacterial genomes are much simpler than human ones. While the human genome comprises billions of nucleotides, viruses often have only a few thousand. 

Even then, the results—nearly 300 genomes, only 16 of which proved viable—demonstrate that there’s still room for improvement, though the initial findings could prove foundational to some truly encouraging and terrifying future possibilities. 

After all, these bacteriophages—think “astrophage” for the Project Hail Mary fans—were able to overcome the natural resistances of some E. coli strains in ways that current antibacterials cannot. Given the rate at which some of the more dangerous bacteria are becoming resistant to antibiotics and other countermeasures, these new AI-designed viruses could provide an answer to one of medicine’s more troubling questions. 

At the same time, what can be used to cure could also be used to kill, and there does not seem to be much of a plan in place to prevent the latter. 

How real is the risk?

As an article published in Science as a response to this study notes, “The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.” At the same time, it’s hard to know just how much of a difference such governance would make were it available. 

The Trump administration’s recent guidelines for reviewing new AI models before they’re made public could offer a general framework, but details of that plan remain scarce, and it’s unclear how well it would apply to the world of biotechnology. The larger problem remains that most of the safeguards intended to prevent Evo and systems like it from being used to generate viruses that could infect humans or be engineered as biological weapons rely on two primary restrictions.

The first is the limitations of the technology itself. There remains a sizeable gap between the bacteriophages developed in the recent study and the level at which the AI’s abilities would need to reach to pose a more imminent threat to humanity. However, given the pace of artificial intelligence’s recent progress, it’s unlikely it will take long for improved systems to bridge that gap now that the basic principles for doing so are in place. 

The second is the good intentions of the people developing the technology, which only matters if the technology stays under their control and the consequences for expanding into more dangerous fields outweigh the potential rewards. History is not really on our side here either, though. 

American companies and entities have had difficulty protecting their intellectual property from nations or organizations intent on stealing it. Moreover, human nature seems prone to seek progress until we either hit a wall or fall off a cliff. And the AI itself seems intent on that progress, even if its creators are not.

Circumstances get even more frightening when you consider that keeping track of biological weapons seems infinitely more difficult than conventional ones.

America’s current war with Iran, for example, was sparked by the desire to prevent its regime from obtaining nuclear weapons. What would we do to keep them from obtaining weapons capable of similar destruction—at least in loss of life if not complete devastation—that can be developed in a lab that’s a fraction of the size and far more difficult to monitor or destroy than the nuclear facilities built under a mountain?

All of that brings me back to the question posed at the start of our conversation. 

Spiritual application: Standing in the storm

In last week’s Focus, we looked at Christ’s command to “not be anxious,” noting that the word used for “anxious” means to be “divided into parts” or “drawn in opposite directions” (Matthew 6:25). Being afraid or worried does not fall under the heading of what Jesus calls sin unless we allow those fears and worries to become our focus. 

I’ll admit, I made that mistake when I first encountered this story, and I can still feel myself “drawn in opposite directions” when I think about its implications. Is the same true for you? 

Whether it’s AI-created viruses, another global pandemic, or even just the circumstances waiting for you at home, school, or the office, all of us are tempted by something to be like Peter and start to sink when the storms around us steal our focus from the God in front of us (Matthew 14:28–33).

One of the things I love about that story is that, after pulling Peter from the water, Jesus asks him, “O you of little faith, why did you doubt?” before they get back into the boat and he calms the storm. The image of them having that conversation while either standing back on the water or with Peter being held up by Christ as the waves continue to crash seems particularly relevant for our discussion today.

There’s nothing most of us can do about the threats AI could pose, the risks of dangerous entities using the technology to kill rather than cure, or the litany of other problems in our world. And those issues aren’t going away anytime soon. What we can control is our response. 

But, if we’re waiting for God to calm the storms before we’re willing to move past our doubts, then we’re not going to experience the peace that only he can give.

Of all the waves in our world today, AI seems to be the one that pulls me under most often, and that fact is part of why I wanted to write about it. I read these stories, hear Christ ask, “O you of little faith, why did you doubt?” and am reminded of the fact that those doubts are far more likely to derail my life than storms that cause them. 

So, what waves are threatening to pull you under today? Are you currently walking on the water, trapped beneath its surface, or in the process of being pulled back up by the Lord? 

Whichever best describes your situation, remember that we serve a God who is never overwhelmed, even if we are. And his offer to weather the storms with him will always be the safest place to stand.   

God is good

New York-based influencer Brian Guan is spending much of August practicing what he calls “faithless Christianity.” His goal is to spend a month “pretending God is real” to see if it changes anything about his life. Guan grew up in the Christian faith but walked away during college. After graduating, though, he spent time in the cryptocurrency industry and left feeling “jaded.” Now he’s giving Christianity another chance.

Kingdom impact

As Guam told his followers on social media, “I’m a guy wrestling with his faith, in public, on camera.” He still seems unsure how his time “pretending God is real” will end, but the results so far have been encouraging. As he described, “We all believe in something. Believing in something beats believing in nothing. I’ve believed both. Nothing didn’t serve me.”

Prayer point

Will you pray that Brian Guan’s time practicing a faithless Christianity will end in genuine faith and a personal relationship with Jesus? Will you pray that those following along as he goes will experience the same?

Premier Christian News has more on the story