Haihtuva: A Data Room That Never Holds Your Files
Haihtuva reads a deal’s documents from the seller’s own storage, on demand, and never makes a copy. The buyer picks which AI model does the reading. When the room closes, the data and the keys are both destroyed. Here is what that changes for confidentiality in a deal, and why BSUM built it this way after 25+ years of hands-on life-science due diligence.
“Old-style data rooms with AI added on look like someone used duct tape to stick the AI on top,” Juhani says. “We built Haihtuva the other way round, from the file up.”
What’s different
Read-only access to the seller’s own repository. Documents stay in the seller’s own Google Drive, OneDrive/SharePoint or Dropbox. Haihtuva reads them on demand and never uploads or copies them anywhere else.
Volatile memory, in an attested computing environment. Each room runs on its own Azure confidential virtual machine in the EU. The machine proves what it is running before it is given any decryption key, holds everything in encrypted volatile memory only, and never writes a document to disk. A single file sits in that memory only for as long as it is actually being read, by a person or by the AI, and is cleared the moment that read ends — the index built across the room’s documents lives longer, for as long as the room stays open, and is destroyed with everything else at close.
The buyer chooses the AI model. Each buyer picks a sealed model inside the room or connects their own model account, with their own prompts, checklists and training material kept private to them, even when several buyers are reviewing the same room at once.
25+ years of hands-on life-science due diligence behind it. Haihtuva is built by BSUM out of what that work showed about where deals actually leak; the platform itself is built for any industry, not only life science.
When the room closes, the data and the keys are destroyed and both sides receive a signed deletion receipt.
In Juhani’s words
What did 25+ years of life-science due diligence teach you about where data rooms actually leak?
Absolutely in the following step: you let the investor or buyer download your data. Simple as that.
Haihtuva’s documents live in encrypted memory and never touch disk. How does a buyer verify the files were actually cleared once a room closes?
Each room runs on its own Azure confidential virtual machine that has to prove what it is running before it is ever given a decryption key. Files sit in encrypted memory only and are never written to disk. When the room closes, the data and the keys are destroyed, and both sides get a signed deletion receipt confirming it — a buyer’s security team can ask to see that receipt and the attestation record.
Why let the buyer, not the seller, choose the AI model that reviews the documents?
Because the buyer is the one who has to trust the output. If the seller or the vendor picked the model, a careful buyer would have to wonder what it was tuned to find, or miss. Letting the buyer bring their own model, their own prompts and their own checklist means the review reflects their judgment, not anyone else’s — and the seller never sees which model was chosen or how it was used.
What are the first paid pilots testing?
A small number of paid pilots, each one applied for at haihtuva.com and approved and onboarded by me personally. They’re testing what matters most before a wider launch: whether the AI surfaces the highest-impact risks instead of noise, whether a buyer’s own model and checklist actually change what gets found, and whether sellers are comfortable granting AI-only or watermarked access to specific folders instead of opening everything.
Try it
BSUM is opening a small number of paid pilots for companies preparing a raise, a licensing deal or an exit, and for investors and acquirers who run due diligence. Apply at haihtuva.com.