Level 1 — Absolute Beginner
Scientists use a computer program called Claude. The program can read and think about information. This kind of program is called artificial intelligence, or AI.
The scientists give the AI a big job. It looks at more than 200,000 proteins. Proteins are very small parts inside living things.
About 950 copies of the program work together. They work for 21 hours. Then they find something new.
They find a new tool inside a virus. The virus is very small and lives inside bacteria. Scientists think this new tool may help them change genes one day.
- scientist
- A person who studies nature and does experiments.
- computer
- A machine that stores information and does calculations.
- program
- A set of instructions that tells a computer what to do.
- protein
- A very small part inside living things that does important work.
- virus
- A very small thing that can make people or other living things sick.
- bacteria
- Very small living things; some are helpful and some make you ill.
- gene
- A small part of a living thing that decides how it grows.
- tool
- Something you use to do a job.
Level 2 — Elementary
The artificial intelligence company Anthropic says its Claude system has found a new enzyme system that scientists had never described before. An enzyme is a protein that speeds up chemical reactions inside living cells.
To make the discovery, the company did not use one AI assistant but about 950 of them at the same time. The agents ran for 21 hours and used 210 million tokens, the units of text that language models read and write. Together they searched through more than 200,000 protein sequences.
The system they found sits inside bacteriophages, which are viruses that attack bacteria. The researchers named it array associated reverse transcriptases, shortened to ART. It combines a reverse transcriptase, an enzyme that copies RNA back into DNA, with repeated pieces of genetic code that look similar to those used in CRISPR.
CRISPR is already used around the world to edit genes, so any system that works in a similar way attracts attention quickly. Scientists warn that the work still has to be tested in a laboratory before anyone can say what ART really does, but the discovery shows how AI can be used to search huge biological databases.
- enzyme
- A protein that speeds up a chemical reaction in a living thing.
- cell
- The smallest unit that living things are made of.
- agent
- A software program that carries out tasks on its own.
- sequence
- An ordered list, for example the order of letters in a piece of DNA.
- bacteriophage
- A virus that infects and kills bacteria.
- DNA
- The molecule that carries genetic instructions in living things.
- RNA
- A molecule that carries instructions copied from DNA.
- database
- A large organised collection of information stored on computers.
Level 3 — Intermediate
Anthropic has reported that a coordinated swarm of roughly 950 Claude agents, working for 21 hours and consuming around 210 million tokens, identified an enzyme system in bacteriophages that had not previously been described in the scientific literature. The agents sifted more than 200,000 protein sequences, a volume that would occupy a human research group for a very long time, and converged on a recurring pattern that nobody had flagged.
The proposed system has been named array associated reverse transcriptases, abbreviated to ART. Its two components are familiar individually but unusual in combination. A reverse transcriptase is an enzyme that copies RNA back into DNA, reversing the normal direction of genetic information flow. The arrays are repeated stretches of genetic code separated by spacers, structurally reminiscent of the repeat and spacer architecture that makes CRISPR systems work as a bacterial immune memory.
That resemblance is what has drawn interest. CRISPR based editing has reshaped biology over the past decade, and any naturally occurring system that stores sequence information in arrays and can write it into DNA is worth examining as a possible alternative tool. It may also tell researchers something about the long arms race between bacteria and the viruses that infect them, since a phage carrying a CRISPR like module is essentially borrowing its host's own defensive grammar.
The important qualification is that computational prediction is not experimental proof. What the agents produced is a strong hypothesis, supported by patterns across many genomes, that now requires wet laboratory work to confirm. The wider significance may lie less in ART itself than in the method: deploying hundreds of reasoning agents against a biological database is a new way of generating candidates for scientists to test.
- swarm
- A large group acting together at the same time.
- converge
- To come together towards the same point or conclusion.
- array
- An ordered arrangement of repeated elements.
- spacer
- A short segment of DNA that separates repeated sequences.
- architecture
- The overall structure and arrangement of a system.
- arms race
- A contest in which each side keeps developing new ways to beat the other.
- hypothesis
- A proposed explanation that still needs to be tested.
- qualification
- A statement that limits or adds a condition to a claim.
Level 4 — Advanced
Anthropic's claim is deliberately narrow and therefore more interesting than the headline suggests. Roughly 950 Claude agents, running in parallel for 21 hours at a cost of some 210 million tokens, worked through upwards of 200,000 protein sequences and surfaced a recurrent architectural motif in bacteriophage genomes that the existing literature had not characterised. The company has named it array associated reverse transcriptases, or ART. What is being asserted is the identification of a pattern worth investigating, not a validated mechanism.
The motif pairs a reverse transcriptase, the enzyme class that writes RNA back into DNA and thereby inverts the canonical flow of genetic information, with arrays of short repeats separated by spacers. That repeat and spacer grammar is the structural signature of CRISPR, which functions in bacteria as an adaptive immune memory: sequences captured from previous infections are stored in the array and used to recognise the invader on return. Finding an analogous arrangement inside the phage rather than the host inverts the usual picture and raises an obvious question about who is defending against whom.
For applied biology the appeal is straightforward. A decade of CRISPR engineering has demonstrated the value of systems that combine addressable sequence memory with enzymatic activity on nucleic acids, and reverse transcriptase based editing already underpins prime editing. A naturally evolved module that unites both functions is, at minimum, a candidate worth characterising, and at best a starting point for a tool with different targeting constraints than the Cas nucleases now in routine use.
The methodological point may outlast the biological one. Large scale sequence mining is not new, and hidden Markov models have been pulling motifs out of genome databases for decades. What is new is the substitution of hundreds of general purpose reasoning agents for a purpose built pipeline, with the agents reading annotations, forming and discarding hypotheses and deciding what to look at next. That is a different kind of search, and its output is a queue of testable propositions rather than an answer. Until the wet laboratory work is done, ART belongs in that queue.
- motif
- A recurring pattern or element, here in a biological sequence.
- canonical
- Accepted as the standard or conventional form.
- adaptive immunity
- Defence that learns from past infections and responds faster on re-exposure.
- analogous
- Similar in function or structure to something else.
- nucleic acid
- A large molecule such as DNA or RNA that carries genetic information.
- nuclease
- An enzyme that cuts nucleic acids such as DNA.
- pipeline
- A fixed sequence of automated processing steps.
- proposition
- A statement put forward for consideration or testing.
Talk about this article with KatePractice this story's vocabulary in a real spoken conversation with your AI tutor