Skip to content
Company Data Advisors

Blog

Can a biotech or R&D company license its lab records to AI buyers?

· Company Data Advisors

Five AI data buyers name science or life sciences, and three describe the records, asking for experiments with failed runs kept. Who owns lab data.

Often yes, if the records are unpublished and the company owns them. Five of the 17 AI data-buyer programs in our buyer index name scientific, research or life-sciences records or industries, and the most detailed, Mercor's science listings, ask for dated lab notebooks and run records that link what was tried to what was measured, with the failed runs kept. The harder question is ownership. Data generated for a sponsor, a client or a partner may belong to them, and Mercor's listings say they don't want data that has already been published.

We read every program's own page on November 14, 2026. Criteria change often, so check the page before you apply.

Which AI data buyers want scientific and R&D records?

Five AI data-buyer programs name scientific, research or life-sciences records or industries. Three describe the records: Mercor in detail, Labelbox Prism and Handshake in a line each. Mercor has a "Science & R&D" category with three listings, and Labelbox Prism marks "Scientific reasoning traces" as urgent on its demand board. Prism defines urgent as "a funded, time-sensitive contract awaiting supply."

Program What it names How specific
Mercor Three listings: Biotech & Life Sciences Lab; Semiconductor & Materials Company; Chem / Mat Science R&D records Records, size, history, tools and exclusions per listing
Labelbox Prism "Scientific reasoning traces" (urgent); "Healthcare and Life Sciences" vertical (high demand) Describes the record type
Handshake "Pharmaceuticals & life sciences: Research documentation, lab protocols, development workflows" Example records only
Scale "Healthcare & Life Sciences" and "Scientific Research" among its industries of interest Industry name only
Centific "Pharmaceuticals & Biotech" in its healthcare card; "Chemicals & Materials" is an option in its form's industry list Industry name; the records it describes are operations records

Sources: each program's own page, read on November 14, 2026. Each buyer's page in our index links to its program.

Prism describes the category as "Scientific reasoning capturing step-by-step expert thought (from diagnostic chains and compound synthesis to circuit design) showing how domain experts reason through complex problems." Handshake lists "Research documentation, lab protocols, development workflows" as the pharmaceutical and life-sciences records it values. Scale names "Scientific Research" as an industry of interest without describing the records, and Centific names "Pharmaceuticals & Biotech" in a card whose examples are claims, care coordination and clinical documentation rather than research.

The other twelve programs don't name science. Some take any industry: SuperAnnotate says "Any industry qualifies. What matters is that the records are specific to your field and hard to source anywhere else," and Polyshares takes "Any business with real operating history, in any sector." Our read: a lab's operating records (email, tickets, finance, project files) should be judged by those programs on the same terms as any other company's.

What do Mercor's science listings ask for?

Mercor's three science listings ask for records of experiments tied to their results. The biotech and semiconductor listings add that the records must be owned, unpublished and cover 1,000 or more runs; the chemistry and materials listing describes the research records it wants in the most detail but states no size, history or exclusions.

Mercor listing (read 2026-11-14) Company size Track record Records it wants Not a fit What the company earns, per Mercor
Biotech & Life Sciences Lab 10 to 1,000 employees Several years of experiment archives Electronic lab notebooks or structured experiment records the organization owns; 1,000+ runs with inputs and measured outcomes, failed runs included Already published data; data owned by a sponsor or partner; companies whose data is mainly software logs $50k to $875k
Semiconductor & Materials Company 10 to 2,000 employees Several years of runs Run or build records with recipe or process settings linked to measured performance; 1,000+ runs, builds, wafers or experiments, failed ones kept with dates Data owned by a customer or foundry partner; published papers and datasheets only; licensed third-party IP $100k to $750k
Chem / Mat Science R&D records Not stated Not stated Linked research episodes: hypotheses, protocols, raw observations, analysis, failures, internal reviews and stop/go decisions Not stated $10k to $300k

Source: Mercor's listings, read on November 14, 2026. Mercor says the amount "Depends on volume and depth of records" on the biotech listing and "Depends on run count and how well runs link to results" on the semiconductor listing.

The biotech listing names the companies it has in mind: "Biotech and biopharma, CROs, genomics and diagnostics labs, consumer genomics companies, academic spin-outs." It lists the systems it works with as Benchling, LabArchives, Dotmatics, LIMS and Medidata Rave. The semiconductor listing names fabs and foundries, device makers, fabless chip designers, equipment makers, and materials and battery R&D labs, and it works with MES, SPC and yield systems, LIMS, electronic lab notebooks, Cadence, Synopsys and OSIsoft PI.

The chemistry and materials listing gives its priority domains as "Materials, chemistry and batteries," "Preclinical biology and biotech," "Computational science and scientific ML," "Industrial engineering and process R&D," "Agriculture, environmental and climate R&D" and "Clinical research." It also says "Records made at the time are preferred over looking back afterwards."

Each listing says Mercor "scopes the data, signs an NDA and runs the export, removing or protecting personal information." For hardware and process records closer to engineering than to the lab, see our guide for engineering and manufacturing companies.

Why do AI buyers want failed experiments?

Mercor's listings say plainly that failed work is wanted, and they ask for the reasoning behind it. Mercor's biotech listing says "Failed experiments are as valuable as successful ones." Its chemistry and materials listing asks for "the reasoning behind the work, not just the scientific facts or the polished papers," and wants "Failed, null, inconclusive, replicated and successful work," "with outcome labels kept." Its semiconductor listing calls "Lab notebooks that explain why a build failed" "especially useful."

The Mercor chemistry listing says the most useful records are "linked research episodes," each of which "connects the question, the evidence available at the time, the competing explanations, the experiments or analyses chosen, the results and the decisions that followed."

Our view: that structure resembles the tasks buyers build into practice environments for AI agents, where an agent attempts a task and is scored against what really happened. A model learning to do research needs to see how researchers chose what to try next, and most of that never reaches a paper. So for a research company, the records that look least presentable are often the most useful. A notebook entry that says why a synthesis was abandoned, linked to the instrument output that showed it failing, is closer to what Mercor and Prism describe than a polished slide deck of results. Our post on flight simulators for office work explains how those environments work.

Published results are the opposite case: both Mercor science listings that state exclusions rule out data that has already been published.

Who owns your lab data?

Holding lab data is not the same as owning it, and research companies often hold data that belongs to someone else. Mercor names the cases directly: its biotech listing excludes "Data owned by a sponsor or partner," and its semiconductor listing excludes "Data owned by a customer or foundry partner" and "Licensed third-party IP." The biotech listing asks "Do you own the data outright, including failed runs?" and the semiconductor listing asks "Do you own the data, and has any of it been published?" These are the cases to sort before applying.

  1. Work done for a sponsor or client. Data generated in a sponsored study is often the sponsor's. One example is a model Accelerated Clinical Trial Agreement between a CRO and a research site for sponsor-funded trials. It says "Sponsor owns and has the right to use the Data," while the site keeps the right to use it for purposes including publication, regulatory and internal research. It also says the study data does not include "research notebooks, source documents, or other routine internal documents," which stay the site's property. Your own contracts decide, so read each one.
  2. Grant-funded research. The NIH Grants Policy Statement says "In general, recipients own the rights in data resulting from a grant-supported project," unless the award's terms say otherwise. It also says NIH "must be given a royalty-free, nonexclusive, and irrevocable license" for federal purposes. Other funders and grant types set their own terms.
  3. Collaborations and licensed-in technology. Joint-development, materials-transfer and technology-license agreements often say who owns results and what can be disclosed. Mercor's semiconductor listing excludes licensed third-party IP outright.
  4. Patient and genomic data. Mercor's biotech listing asks "Is any of it patient-level or genomic data, and under what consent?" and Prism says multimodal medical records need "IRB approval or qualified de-identification." Our guide for medical practices covers HIPAA de-identification, and the consent forms your subjects signed may limit reuse further.
  5. Your lab software. Check the vendor's terms. Benchling's Main Services Agreement says the customer "exclusively owns and reserves all right, title and interest in and to Customer Data."

Our view: unpublished results may also matter for patents and trade secrets. If you might still file on work in the records, or the records describe a process you keep secret, talk to patent counsel before any record leaves the company, even under an NDA. We'd also ask counsel whether any semiconductor, defense or chemistry records fall under export-control rules before they go in any package. We flag these questions; we don't give legal advice.

What if the company is winding down?

A wound-down biotech or hardware startup can still qualify. Mercor's biotech listing says "Biotechs that have wound down and labs with years of archived work are a strong fit," and its semiconductor listing says "Labs and startups that have wound down qualify." The risk is losing the records before anyone licenses them. Benchling's agreement says that "After sixty (60) days following termination or expiration of all Sales Orders, Benchling may, but is not bound to, irretrievably erase" customer data. Other vendors have their own terms; check each one before you cancel a subscription.

Before you shut systems down, export the electronic lab notebooks, LIMS and instrument data together, so runs stay linked to their results, and keep the failed runs with their outcome labels. Our guide to licensing when you're closing the business covers who can sign once a company is winding down, which matters when a lender, an assignee or a board holds the decision.

Where we come in

Five programs name scientific or life-sciences records or industries today, and they describe what they want at very different levels of detail, with exclusions (sponsor data, partner data, published data, licensed IP) that decide fit before price does. We track every program's published criteria and terms, have direct contacts at the major data buyers and at AI labs, and compare the programs that fit your records on the whole deal: price, exclusivity, how long the license runs, deletion and onward use. Nothing about your company is shared until you approve a specific, named buyer. Participating buyers may pay us referral fees, and we tell you which before any introduction. Owners pay us nothing, and you can apply to any buyer directly.

What to check before you apply

Answer these six questions first. They are the ones the buyers' own listings ask.

  1. Which systems hold your experiment records (ELN, LIMS, instrument software, shared drives), and how many years do they cover?
  2. Roughly how many runs, builds or experiments do you have, and do the records link each one's inputs to its measured result?
  3. Did you keep the failed and repeated runs, with dates and outcome labels?
  4. Which records were made for a sponsor, client, customer or partner, or under a grant or collaboration agreement, and what do those contracts say about ownership and disclosure?
  5. What has already been published, and what is still unpublished?
  6. Is any of it patient-level or genomic data, and what did the consent cover?

Then compare your answers with the tables above, or check where you fit. For what buyers look for beyond the science itself, see is your company's data worth anything to AI companies.

Programs in this post

See which buyers fit your company

Eight quick questions against every program in our buyer index. Participating buyers may pay us referral fees.