Datasets:

aitrainingdatabroker
/

ai-data-deal-checklists

id stringlengths 7 7	checklist stringclasses 4 values	section stringlengths 4 18	kind stringclasses 3 values	audience stringclasses 2 values	item stringlengths 32 112	why stringlengths 24 125
chk-001	check-an-offer	Terms to settle	check	seller	Is it a real offer or an estimate?	An estimate is a range from a few questions. An offer is one amount in writing, with the terms attached and a name behind it.
chk-002	check-an-offer	Terms to settle	check	seller	Is any exclusivity paid for and limited?	An exclusive AI lab agreement with no end and no extra payment can stop you from licensing the same data to anyone else.
chk-003	check-an-offer	Terms to settle	check	seller	Does the license say how long it lasts and what happens after?	A trained model keeps what it learned after a license ends, so ask about copies, passing the data on, and deletion.
chk-004	check-an-offer	Terms to settle	check	seller	Is it written down who removes personal information, how, and who pays?	It should happen before the data leaves your control, usually paid by the AI lab, with a cleaned sample you can review.
chk-005	check-an-offer	Terms to settle	check	seller	Are the payment timing and method clear?	Some offers pay at delivery, some over time, some only as the data is licensed on. Ask what is guaranteed.
chk-006	check-an-offer	Terms to settle	check	seller	Is it clear what can be taken back?	Look for rights to claw money back if data is judged incomplete, part is rejected, or a promised right is not yours.
chk-007	check-an-offer	Terms to settle	check	seller	Do you know who pays the person in the middle?	Intermediaries are paid by the buyer, by a share of what you receive, or by a fee from you. Know which before signing.
chk-008	check-an-offer	Signs to slow down	warning	seller	You are pushed to sign before reading the whole license.	Pressure to decide fast is a reason to slow down.
chk-009	check-an-offer	Signs to slow down	warning	seller	The buyer wants standing access to your live systems with no end date.	An export or temporary read-only access is safer.
chk-010	check-an-offer	Signs to slow down	warning	seller	No cleaned sample is shown to you before it goes anywhere.	You should review what leaves your control.
chk-011	check-an-offer	Signs to slow down	warning	seller	Nobody will say how personal information is removed, or you are told there is zero risk.	Cleaning lowers risk but does not remove it; a clear method and a reviewable sample matter.
chk-012	check-an-offer	Signs to slow down	warning	seller	An exclusive agreement is requested with no time limit and no extra payment.	Sole rights are worth paying for and should be limited.
chk-013	check-an-offer	Questions to ask	question	seller	Is this one amount in writing, or a range?	Separates an offer from an estimate.
chk-014	check-an-offer	Questions to ask	question	seller	What exactly am I licensing, and what do you keep?	Defines scope and your remaining rights.
chk-015	check-an-offer	Questions to ask	question	seller	May you sell it on or share it with others?	Passing data on changes its value and your risk.
chk-016	check-an-offer	Questions to ask	question	seller	Who removes private information, when, and can I see a sample first?	Puts cleaning in writing before any delivery.
chk-017	check-an-offer	Questions to ask	question	seller	What is guaranteed, and when would I be paid?	Separates promised money from possible money.
chk-018	check-an-offer	Questions to ask	question	seller	What can you take back, and under what conditions?	Reveals clawback rights.
chk-019	de-identification	Before you start	check	seller	List where personal information can sit: names, emails, phones, addresses, account and order numbers, free text.	You cannot remove what you have not found.
chk-020	de-identification	Before you start	check	seller	Decide which records to leave out altogether.	Some records are better excluded than cleaned, such as health or payment details.
chk-021	de-identification	Before you start	check	seller	Check your customer terms and privacy policy for what you may share.	Rights and consent come before cleaning.
chk-022	de-identification	Clean the data	check	seller	Replace each person with a consistent placeholder, such as Customer 1.	A consistent placeholder keeps who-said-what readable.
chk-023	de-identification	Clean the data	check	seller	Replace emails, phones and ID numbers with labeled placeholders.	A label such as [EMAIL] keeps the structure of the text.
chk-024	de-identification	Clean the data	check	seller	Generalise addresses to a region and dates to a season and year.	Exact places and dates can point to a person.
chk-025	de-identification	Clean the data	check	seller	Round amounts where exact figures could identify a deal or person.	Precise figures can act as fingerprints.
chk-026	de-identification	Clean the data	check	seller	Look for indirect identifiers: rare job titles, small towns, unusual events.	Combinations of ordinary details can still identify someone.
chk-027	de-identification	Clean the data	check	seller	Clean attachments, signatures and metadata, not only message bodies.	Personal details often sit in footers, file properties and quoted replies.
chk-028	de-identification	Review	check	seller	Review a random sample of the cleaned records by hand.	Automated cleaning misses cases; a human check finds them.
chk-029	de-identification	Review	check	seller	Test the cleaned set with a second method or tool.	A different method catches different leaks.
chk-030	de-identification	Review	check	seller	Record what was removed and how.	A written method is what counsel and buyers ask to see.
chk-031	de-identification	Sign-off	check	seller	Have your counsel review the method against the rules that apply.	Examples are HIPAA, CCPA and CPRA, GDPR and UK GDPR, and Canada's PIPEDA and Quebec Law 25. This is not legal advice.
chk-032	de-identification	Sign-off	check	seller	Where a rule requires a qualified expert to sign off, use one.	Some frameworks name a method that needs an expert's determination.
chk-033	de-identification	Sign-off	check	seller	Approve the cleaned sample before it leaves.	You make the final call on what is shared.
chk-034	seller-readiness	What you hold	check	seller	Name the kinds of records you hold and the years they cover.	Years of connected records matter more than size.
chk-035	seller-readiness	What you hold	check	seller	Check that each task is followed through to an outcome.	Outcomes kept next to the work make data more useful.
chk-036	seller-readiness	What you hold	check	seller	Estimate how many records and how many customers or cases.	Variety and volume are both part of a description.
chk-037	seller-readiness	Rights	check	seller	Confirm who created the data, including staff and contractors.	Creators may hold rights you need to clear.
chk-038	seller-readiness	Rights	check	seller	Read customer terms and contracts for limits on reuse.	Some forbid reuse or require consent.
chk-039	seller-readiness	Rights	check	seller	Separate third-party content mixed into your records.	You cannot license what you do not hold.
chk-040	seller-readiness	Export	check	seller	Confirm you can export the data in a standard format.	Data that cannot be exported cannot be delivered.
chk-041	seller-readiness	Export	check	seller	Keep a copy before cancelling any system.	Records can disappear when a subscription ends.
chk-042	seller-readiness	Deal	check	seller	Decide whether you could grant any exclusivity.	Exclusivity changes price and options.
chk-043	seller-readiness	Deal	check	seller	Tell your broker if you already contacted an AI lab and where it stands.	It avoids two parties presenting the same data to the same lab.
chk-044	seller-readiness	Deal	check	seller	Plan who on your team reviews the sample and signs off.	Reviews are the slowest step when nobody owns them.
chk-045	source-company-data	Brief	check	buyer	State the task type you want the data to teach.	A specific brief gets better matches.
chk-046	source-company-data	Brief	check	buyer	State the years of history and approximate volume you need.	Depth over time is often the scarce part.
chk-047	source-company-data	Brief	check	buyer	State the format and the outcome fields required.	Outcomes kept next to the work decide usefulness.
chk-048	source-company-data	Brief	check	buyer	State the exclusivity and license term you can accept.	Sellers price exclusivity and duration.
chk-049	source-company-data	Seller diligence	question	buyer	Who created the data, and what do the customer terms say?	A clear chain of rights lowers risk.
chk-050	source-company-data	Seller diligence	question	buyer	How was personal information handled, by whom, and can we review a sample?	Privacy handling is part of the product.
chk-051	source-company-data	Seller diligence	question	buyer	How was the data collected and how complete is it?	Gaps and collection method affect training value.
chk-052	source-company-data	Seller diligence	question	buyer	What can the seller confirm, and what can it not?	Honest limits are a good sign.
chk-053	source-company-data	Contract	check	buyer	Agree who pays for cleaning and review before delivery.	Usually the AI lab, decided case by case.
chk-054	source-company-data	Contract	check	buyer	Forbid attempts to identify any person in the data.	A standard clause in licensed company data.
chk-055	source-company-data	Contract	check	buyer	Write down deletion and copy terms for the end of the license.	Trained models keep what they learned, so agree what happens to the data itself.

AI Data Deal Checklists

Four checklists for company data deals with AI labs, one item per row so they can be filtered, counted, turned into forms or fed to a model:

Checklist	Rows	For
`check-an-offer`	18	a seller reading an offer: terms to settle, signs to slow down, questions to ask
`de-identification`	15	a seller cleaning data: before you start, clean, review, sign-off
`seller-readiness`	11	a seller before approaching a buyer: what you hold, rights, export, deal
`source-company-data`	11	an AI lab or model team sourcing licensed data: brief, seller diligence, contract

Fields

Field	Meaning
`id`	`chk-001` to `chk-055`
`checklist`	which checklist the item belongs to
`section`	the heading inside the checklist
`kind`	`check`, `question` or `warning`
`audience`	`seller` or `buyer`
`item`	the thing to check, ask or watch for
`why`	one sentence on why it matters

Load it

from datasets import load_dataset
ds = load_dataset("aitrainingdatabroker/ai-data-deal-checklists", split="train")
offer = ds.filter(lambda r: r["checklist"] == "check-an-offer")

Limits

One broker's view, written in October 2026. Market practice and rules change.
Laws are named only as examples of what to ask your counsel about. This is not legal advice and not a certification; where a rule requires a qualified expert, use one.
No prices and no buyers: nothing here is an offer.

build.py rebuilds data/train.jsonl from the source list.

License and credit

CC BY 4.0. Please credit AI Training Data Broker and link to https://aitrainingdatabroker.com/de-identification/ when you share or build on it.

About

AI Training Data Broker is a broker and technical advisor for companies that sell or license their data to AI labs. It guides the technical side, including export, format, de-identification and compliance, and publishes free checklists on de-identifying data and keeping it within the privacy rules that apply.

Downloads last month: 33

Number of rows:

Total file size:

30.5 kB