Anthropic’s biology research lab says it used Claude at every step of a project that produced the first 3D structure of a vault particle found in prokaryotic cells. In a research post published on 4 November, it says “a vault purification that conventionally takes three days took us about 90 minutes”. That is about 48 times faster, by our arithmetic (72 hours against 90 minutes).
The same post says “a first structural model was ready 34 hours later, which we estimate would normally have taken us weeks”. Anthropic, which sells Claude to research labs, lists the misses alongside the gains in the same post.
Business Pill · The Model and the Harness
An evergreen Business Engineer explainer on the model and the harness around it. It teaches the general idea only and does not discuss Anthropic’s vault research.
The key insight: Anthropic says its biology lab used Claude at every step, from a search of about 15 billion predicted proteins to processing the microscopy data, and that an isolation step fell from about three days to about 90 minutes, about 48 times faster by our arithmetic; the same post lists a protocol error that cost two weeks.
Anthropic describes the division of labour this way: “Under our direction, Claude searched protein databases, designed the DNA constructs, optimized the isolation method, and processed the microscopy data.”
The search ran on Claude Science. The post says the team “began by using Claude Science to search for the vault protein’s relatives across a database of about 15 billion predicted proteins”. The relatives turned out to be rare: present in 29 bacterial phyla and four of archaea, but in less than 1% of species with a sequenced genome.
From there the post describes a narrowing funnel. The team chose 44 proteins to test; in E. coli, “15 of 24 produced soluble protein”; “seven of the 15 proteins had assembled into vault-like particles”; and cryo-EM confirmed the vault shape for four of those seven. The detailed structure is of one phage protein, L05, which the post says forms a vault from 96 copies of a single protein, against 78 in human vaults.
On the DNA work, the split was explicit: “We chose the proteins, and Claude turned each protein sequence into DNA tailored to the host cell”. The team also made “15 altered versions of the protein” working with Claude, to learn more about how the proteins assemble; the post says some single changes near the cap were enough to prevent the shell from closing.
Vault purification time as stated by Anthropic in its 4 November 2026 research post: a step that conventionally takes about three days took about 90 minutes with the method Anthropic’s team optimised with Claude. These are Anthropic’s own figures; the ~48x multiple (72 hours vs 90 minutes) is our arithmetic.
What Was Posted on X
Weโre sharing how our biology research team worked with Claude to map the first 3D structure of a vault particle from bacterial life.
Vault particles are a longstanding mystery. They are barrel-shaped objects, among the largest protein structures found in our cells. They were first discovered by Leonard Rome and Nancy Kedersha in 1986, but their function remains unknown.
With Claude, weโve found results that might bring us closer to solving that mystery.
Until now, vaults were only known in the cells of complex things like animals and plants. We have shown that similar proteins from bacteria (and the viruses that infect them) can also assemble into vaults. Finding the structures in these simpler forms of life may make it easier to investigate their function in our own cells.
For our life sciences team, the project was also a demonstration of the potential (and the limitations) of using AI to accelerate science.
In our full piece, we discuss how Claude was able to dramatically speed up some stagesโmanually determining the vault particleโs structure would have taken weeks to months of painstaking work, but here it took days. We also discuss how Claude made some simple mistakes along the way, and how we collaborated with our machine learning researchers to address them.
Anthropic’s post on X of 4 November announcing that its biology research team worked with Claude to map the first 3D structure of a vault particle from bacterial life. It says Claude dramatically sped up some stages and also made some simple mistakes along the way. The piece uses Anthropic’s linked research post for the figures.
This is INSANE: Anthropic says Claude helped uncover the FIRST 3D structure of a vault-like particle made in bacteria, shedding light on a biological mystery dating back 40 YEARS.
Claude searched 15 BILLION protein sequences, helped cut purification from THREE DAYS to just 90 MINUTES, and produced an initial structural model in 34 HOURS instead of weeks.
@choblin29 posted this at 20:22 UTC on 4 November 2026, saying that Anthropic says Claude helped uncover the first 3D structure of a vault-like particle made in bacteria, a mystery the post dates back 40 years. The poster adds that Claude searched 15 billion protein sequences, helped cut purification from three days to 90 minutes and produced an initial structural model in 34 hours instead of weeks.
Where Claude Pushed Back, and Where It Slipped
The 90-minute step came from a method Claude first advised against. The post says: “Claude first argued against taking the chromatography approach, because clumps of misfolded protein come out in that same void volume.” The team went ahead anyway, and the post says Claude then helped make it work, including a suggestion the scientists were “reluctant at first” to try because the columns are expensive.
Anthropic also lists the errors. One protocol Claude wrote “set the project back by two weeks”. Early in the image processing, it “skipped a standard step and produced a flawed map blended with its mirror image”, and the team told it to stop improvising: “Please let’s do normal things unless they don’t work”.
The structure work itself ran inside Anthropic: the post says Claude worked “in an internal multi-agent harness with access to internal compute resources”, and that the cryo-EM analysis “used hundreds of GPUs in parallel to save time”.
The Product Claim
Anthropic makes a direct claim about what outside labs could do: “Most of this work could be achieved by a typical lab with access to existing Claude products.” The exception it names is the cryo-EM analysis, which “required more computing resources than might be available to an academic institution”.
The post also describes a loop back into the model. The lab says it “shared our lab notes with the ML researchers who train Claude, so they could see where the model went wrong”, and that “This has informed the kind of data Claude learns from.”
In its post on X, Anthropic called the project “a demonstration of the potential (and the limitations) of using AI to accelerate science”.
What Dr Rome Said
Leonard Rome, who co-discovered vaults in 1986 and advised Anthropic on the research, said when shown the structure: “Figuring out the function of vaults is still enormously challenging. I am pleased that Anthropic has so much good data. Now, other people will be working on this problem … The broadness of Anthropic’s finding[s] suggests that it is going to be a complicated function and an interesting ride.”
He added: “Will we find out that they are all doing the same thing? Or will we see an evolutionary development of vault function? Maybe they are doing the same thing as human vaults, and that is also very interesting.” Anthropic’s disclosure says Dr. Rome was a paid consultant, providing feedback on the manuscript.
As we read the post, the speed-up came from a method Claude first argued against: the team overruled the model, and the model then helped make the method work.
As we read it, the lab also works as a feedback channel: Anthropic says its lab notes went to the researchers who train Claude and informed the kind of data Claude learns from.
Anthropic, Investigating vault particles with Claude, 4 November 2026
“Most of this work could be achieved by a typical lab with access to existing Claude products.”
Three Implications
FOR OUTSIDE LABS Anthropic says most of the work could be done by a typical lab with existing Claude products; the exception it names is the cryo-EM analysis, which used hundreds of GPUs in parallel.
FOR THE MODEL Anthropic says it shared its lab notes with the researchers who train Claude, and that this has informed the kind of data Claude learns from.
FOR THE SCIENCE The vaults’ function is still unknown, the vaults have so far been made only in E. coli, and the preprint has been submitted to bioRxiv.
The framework says: “If you understand how each data pipeline becomes a moat — and how each moat decays — you understand the strategic structure of the entire AI stack.”
As we read it, Anthropic’s post describes one such pipeline: its own lab runs Claude on real experiments, records where the model went wrong, and passes those notes to the people who train it. The framework does not discuss this project.
The vaults’ biological function is still unknown; the post says the work “only deepens the mystery” and that the vaults have so far been made only in E. coli. The speed figures are Anthropic’s own, and the 34-hour comparison rests on its estimate of “weeks”. The preprint has been submitted to bioRxiv, which means it has not yet been peer reviewed. The 48 times figure is our arithmetic on the three days and 90 minutes Anthropic states.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Anthropic’s lab says Claude searched about 15 billion predicted proteins, designed the DNA constructs and processed the microscopy data for the first prokaryotic vault structure. It also helped optimise an isolation step from about three days to about 90 minutes, about 48 times faster by our arithmetic. One protocol error it wrote set the project back by two weeks.
A note on sourcing. Quotations, figures and the steps attributed to Claude are from Anthropic’s research post of 4 November 2026 and its post on X; they are the company’s own account, and the preprint has not been peer reviewed. The 48 times multiple is our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
