0
0
Picture a library with billions of books, and nobody has read most of them. Now picture sending in about 950 tireless readers at once, and telling them to look for one strange pattern.
That is roughly what happened in late September 2026. The AI company Anthropic set about 950 copies of its AI model, Claude, loose on a huge database of DNA. After about 21 hours of searching, they flagged something nobody had named before: a CRISPR-like system hiding in the DNA of viruses.
The headlines were loud. The honest answer to the obvious question is much quieter. What does this new system do? Right now, nobody knows.
And that quiet answer turns out to be the most interesting part of the story.
What Is the CRISPR-Like System the AI Found?
The new system has been named ART, short for array-associated reverse transcriptases. That is a mouthful, so here is the simple version.
Dimitri Perrin, a computer scientist at Queensland University of Technology, explained it as three parts sitting side by side in the DNA:
- An enzyme called a reverse transcriptase. Its job is to copy RNA back into DNA.
- A partner gene that sits right next to that enzyme.
- A long row of repeating DNA, with the same short piece showing up again and again at even spaces.
That last part is why people said “CRISPR.” CRISPR is the famous gene-editing tool, and in nature it also has a long row of evenly spaced repeats. In bacteria, those repeats work like a memory bank. They hold little copies of old enemies, so the cell can spot them next time.
ART was found in the DNA of viruses that infect bacteria. Those viruses are called phages.
How 950 AI Agents Searched Viral DNA in 21 Hours
The team told the AI agents to search for proteins that might work together with reverse transcriptase enzymes. The tech news site The Decoder reported that the search covered more than 200,000 of these enzymes.
No human could read that much in 21 hours. That is the real skill on show here. A machine can sift a mountain very fast and point at the one odd rock.
Stanley Qi of Stanford University told Al Jazeera that what stood out was the AI’s ability to spot an unusual pattern that had been hard to find before. MIT’s Feng Zhang, one of the pioneers of CRISPR gene editing, called the find intriguing and worth a closer look.
Why Some CRISPR Scientists Aren’t Impressed
Not everyone cheered. And the doubters make a fair point.
Kevin Blake, a microbiologist at Washington University, told Al Jazeera: “There’s nothing to indicate this is a rival to CRISPR-the-technology, or could be developed into any kind of therapeutic or practical application.”
Lucas Harrington, a co-founder of the CRISPR company Mammoth Biosciences, went further. He said on X that this kind of DNA search, often called genome mining, has been done for decades. In his view, finding a pattern is the easy part. The hard part is working out what a system actually does, and that has not been shown yet.
Even Anthropic’s own chief executive, Dario Amodei, has said the system’s exact job, and whether it will ever be useful, is not yet clear.
Perrin put it most simply: “At this point in time, we can say that ART is CRISPR-like in its architecture, but there is no evidence that it is CRISPR-like in its function.”
In plain words: it is shaped like CRISPR. Nobody knows yet if it acts like CRISPR.
CRISPR Itself Sat Unexplained for 20 Years
Here is the part most of the coverage skipped. The critics are right that a shape is not a purpose. But the history of CRISPR shows that finding the shape first is normal. It is how this kind of science has always gone.
In 1987, a Japanese team led by Yoshizumi Ishino was studying a gene in E. coli bacteria. By accident, they copied a strange stretch of DNA next to it: short pieces that repeated again and again, with other bits in between. They noted it in their paper. They could not say what it was for.
Then, for years, almost nothing. According to a history of CRISPR written by Ishino and colleagues, until 2007 that first paper was cited only once or twice a year.
The answer finally came in 2007, and it came from an unlikely place: a bacterium used to make yogurt and cheese. Researchers including Rodolphe Barrangou and Philippe Horvath showed that Streptococcus thermophilus stores bits of virus DNA in those repeats. When the bacterium gains a new piece, it gains protection from that virus. The strange pattern was an immune system.
Twenty years passed between seeing the shape and knowing its purpose. And only after that came the gene-editing tool that won Jennifer Doudna and Emmanuelle Charpentier the Nobel Prize in Chemistry in 2020.
So ART may turn out to be a big deal. It may turn out to be a small one. Nobody can know yet, and anyone who tells you for certain is guessing.
Knowing the Shape Before the Purpose
There is something freeing in that gap, if you let it be.
We live in a time that loves quick answers. A tool found it, so surely we understand it. But the living world keeps showing us the same thing. We can map a pattern long before we know what it is for. We can hold a thing in our hands and still not see its reason.
Long before anyone knew what a gene was, an old book of wisdom put it like this. You do not know how bones grow inside a mother’s womb. In the same way, it said, you do not know the full work of God, who makes all things. That was not written to stop anyone from searching. It was written to keep the searcher humble, and a little bit amazed.
A faster search does not shrink the mystery. It just shows us how much more of it there is.
What Happens Next With ART
Now comes the slow part, the lab work. Scientists will need to test what ART actually does inside real cells and real viruses. That could take months. It could take years.
Meanwhile, you can use this story as a simple filter the next time a headline says an AI “discovered” something. Ask two questions. What exactly did it find? And do we know what it does yet?
If the answer to the second question is “not yet,” that is not a letdown. It is the start of the best part. Somewhere out there, a pattern is waiting for someone patient enough to find out why it is there. You can see the same kind of wonder in the largest survey of physicists ever done, which found the frontier still wide open. And if you are wondering what these AI systems are like on the inside, researchers recently found a signal inside AI models that acts a little like pain.
The library is still mostly unread. That is good news.
What Do You Think?
When an AI finds something in nature that people missed, who deserves the credit: the AI, the people who built and aimed it, or the scientists who will spend years working out what it means? Tell us where you land in the comments.
Share This
“950 AI agents searched viral DNA for 21 hours and found a CRISPR-like system. Nobody knows what it does yet. CRISPR’s own repeats waited 20 years for that answer.” https://bgodinspired.com/?p=116849
“Shaped like CRISPR is not the same as acting like CRISPR. This is the calmest, clearest explainer I’ve read on the AI ‘CRISPR-like’ discovery.” https://bgodinspired.com/?p=116849
“Fun fact: the answer to what CRISPR does came from a yogurt bacterium, 20 years after the pattern was first spotted.” https://bgodinspired.com/?p=116849
Common Questions About the CRISPR-Like System
What is the CRISPR-like system that Claude AI found?
It is a set of genes called ART, short for array-associated reverse transcriptases, found in the DNA of viruses that infect bacteria. It has three parts: a reverse transcriptase enzyme that copies RNA into DNA, a partner gene next to it, and a long row of evenly spaced repeating DNA. That repeating row looks like the repeat arrays in natural CRISPR systems, which is why it was called CRISPR-like.
Does ART work like CRISPR gene editing?
Nobody knows yet. Computer scientist Dimitri Perrin said ART is CRISPR-like in its architecture, but there is no evidence that it is CRISPR-like in its function. Its job inside the virus has not been worked out, and there is no evidence so far that it could be used as a gene-editing tool.
How did the AI find the new DNA system?
Anthropic ran roughly 950 AI agents of its Claude model for about 21 hours. They searched large databases of DNA and protein sequences for genes that might work together with reverse transcriptase enzymes. The search flagged an unusual pattern of repeats sitting next to one of these enzymes in the DNA of bacteria-infecting viruses.
Why do some scientists say the AI discovery is not a big deal?
Some CRISPR researchers say searching DNA databases for new gene patterns, called genome mining, has been done for decades. They argue that spotting a pattern is the easy part, and the hard part is proving what the system does. Microbiologist Kevin Blake said nothing so far suggests ART could rival CRISPR technology or become a practical treatment.
How long did it take scientists to learn what CRISPR does?
About 20 years. A Japanese team led by Yoshizumi Ishino first noticed the strange repeating DNA in E. coli bacteria in 1987. In 2007, researchers studying Streptococcus thermophilus, a bacterium used to make yogurt and cheese, showed that the repeats are part of an immune system that helps bacteria remember and fight viruses.
