Software Companies Licensing Code and Product Data to AI

A software company can license its own code, engineering history, product databases and AI-product usage logs to AI companies, but customer content held in its product is governed by the customer terms and privacy promises the company has already made. Mercor states USD 50,000 to 500,000 for code, USD 55,000 to 350,000 for data stacks and USD 20,000 to 480,000 for AI-product sessions, checked 2026-11-09. Four transactions on the public record involve software companies as suppliers, and none discloses a license fee.

Handshake AI, HUMXN, Mercor and micro1 name software companies on their own pages, and DataPacts takes no cut of any sale. Software records are one sector of AI training data licensing.

DataPacts Editorial Team checked the buyer pages, vendor terms, rules and filings cited here on 2026-11-09 and 2026-11-10, following how we source, date and check every figure.

What Software Companies Hold That AI Buyers Ask For

Software companies hold four record families that AI buyers ask for by name, plus a fifth, customer content, that sits in their systems without being theirs alone. Mercor's listings ask for private repositories, production databases and complete session logs, and micro1 states that it wants engineering documentation, bug tracking and agent trajectories with company approval.

The table compares the record families, who asks for each and the first limit, as of 2026-11-10; the Mercor listings were re-read 2026-11-10, other statements checked 2026-11-09.

Session logs are licensed as computer-use and agent trajectory data when an agent acts inside software.

Buyer Programs That Name Software Companies

Four programs name software companies on their own pages, and one of them, Mercor, runs three separate listings for code, data stacks and AI products. The table compares what each program states for software companies, in alphabetical order, as of 2026-11-10; each value is the party's own statement, checked 2026-11-09 unless marked.

Each figure is the stating party's claim with its check date; a program missing here may still accept software companies.

Mercor's program page lists eight target industries and software is not among them, while its listings feed carries three software listings, so both statements are shown.

Acquirers of code alone and their rates are covered under selling or licensing source code to AI companies, so this page repeats none of them.

Clearing a Software Company's Own Code Before It Is Licensed

Buyers take only code the company owns outright, so the first check is whether every contributor's work belongs to the company. Mercor states that its Software Company listing needs private repositories that have never been public, 3+ years of commit history from at least 10 engineers and 50,000+ owned lines (re-read 2026-11-10). Mercor lists four cases as not a fit:

• Forks or copies of open-source projects
• Student, hackathon or side projects, or code mostly written by AI tools
• Client work an agency does not own
• Repositories with a single author

Employers own employee code under 17 U.S.C. 201(b) unless a signed written instrument says otherwise, and under CDPA 1988 s.11(2) and UrhG s.69b unless agreed otherwise, while contractor and freelancer code needs a written assignment or express grant, so employee, contractor and founder assignments decide who owns the code you want to sell.

micro1's terms of 2026-11-08 exclude export-controlled information unless the Order Form allows it, "among other categories".

Controlled source code raises its own question: under 15 CFR 734.13 (81 FR 35603, 2016), releasing controlled technology or source code to a foreign person in the United States is a deemed export to that person's most recent country of citizenship or permanent residency. No export classification of any software company's code is stated here.

Customer Content Held in the Product: Where a Software Company's Rights Stop

Buyers themselves draw the line at customer content: Mercor's SaaS listing rules out agencies holding their clients' databases, its AI Product listing asks for user terms that allow licensing de-identified usage data or a way to get consent, and its Cybersecurity listing takes customer data only with customer agreement. The first two were re-read on 2026-11-10, the third checked on 2026-11-09.

Vendors' own terms show what a company can bind itself to. Slack states that generative AI training on Customer Data needs affirmative opt-in consent, Zoom states in ToS s.10.2 (effective 2023-08-11) that it does not use communications-like Customer Content to train AI models, Google Workspace states that it does not train models on customer data without prior permission or instruction, and Figma's content training, effective 2024-08-15, is off for Organization and Enterprise plans, all checked 2026-11-09. These are examples, not advice about those vendors.

Integration products inherit the terms of the platforms they connect to. Slack API Terms, effective 2025-10-10, bar third-party apps from using API Data to train a large language model, and the Google Workspace API user data policy prohibits using user data to train an AI model beyond a user's personalized model, checked 2026-11-09. Platform terms are one layer of a rights audit.

FTC staff wrote on 2024-02-13 that a surreptitious, retroactive change to terms or a privacy policy to allow AI training may be unfair or deceptive under FTC Act s.5 (15 U.S.C. 45), and on 2024-01-09 that the FTC has required firms that unlawfully obtained data to delete models and algorithms. Oregon DOJ guidance of December 2024 under the OCPA (ORS 646A.570 to 646A.589) requires disclosure of AI training in the privacy notice and no retroactive notice changes, with penalties up to USD 7,500 per violation. Connecticut P.A. 25-113 requires from 2026-07-01 that privacy notices say whether the controller sells or uses personal data to train large language models. Four limits follow:

• The customer terms and privacy promises the company has already made
• The API terms of every platform the product connects to
• Notice and consent rules for personal data used in AI training
• HIPAA, where covered entities and business associates may not sell PHI without authorization (45 CFR 164.502(a)(5)(ii)) and information de-identified under 164.514(a) and (b) falls outside the Privacy Rule (eCFR current as of 2026-11-05)

In the EU, GDPR Art. 6(4) weighs compatibility of purposes before further use and Art. 13(3) requires information before processing for another purpose, applicable since 2018-05-25. Chapter VI of the Data Act (Regulation (EU) 2023/2854, applicable since 2025-09-12) helps cloud and SaaS customers export data when switching provider but is not a general right to export for sale. Legal basis and notices are covered under GDPR and AI training data.

Software Company Deals on the Public Record

Four transactions on record involve software companies as the party supplying data, and none of them discloses a license fee. The table lists them by date, as of 2026-11-09, with values as each party states them.

Two patterns appear: Stack Overflow and Automattic give access to community content, and Automattic states that opt-outs are honored, while Cursor's agreement is an option agreement covering developer interaction data, with fees only on defined termination, never as a data price. The tracker of AI data licensing deals lists every sector's deals with parties and sources.

SimpleClosure, an intermediary and not a buyer, charges a percentage only on success, and its blog of 2026-09-03 states that nearly 500 million lines were offered and roughly 37 % of raw code was removed in cleaning (checked 2026-11-09). It states that it has helped facilitate deals in the tens to hundreds of thousands of dollars for unique assets, not a software-specific price.

Benefits, Risks, Downsides and Alternatives for a Software Company

Licensing can earn income from code and records a software company already keeps, but customer promises, export rules and irrevocable license terms carry risks a fee does not offset, and not licensing is a valid outcome.

Benefits

• Income from records the company already keeps.
• Demand stated by several programs, as claims with check dates.
• A nonexclusive grant, as in micro1's s.1.2, leaves the company free to use its records.

Risks

• Breaching the company's own customer terms or notices (FTC staff, 2024-02-13).
• A deemed export of controlled source code (15 CFR 734.13).
• Security incidents at buyers: Mercor reported unauthorized access through compromised LiteLLM versions, breach dates 2026-03-24 and 2026-03-30 (California AG); lead case N.D. Cal. No. 3:26-cv-02821-WHO.
• An irrevocable, perpetual license with no duty to retrain (micro1 s.1.2 and s.4.4, updated 2026-11-08).

Downsides

• De-identification and consent work, since logs that cannot be stripped of personal information are not a fit.
• Size and stack thresholds, such as 15 to 500 employees for Mercor's code listing and 30+ at micro1.
• Delivered records stay in trained models, since micro1 states no duty to retrain.

Alternatives

• License engineering history only.
• Keep the records for in-house use.
• Sell the code in a shutdown through an intermediary.
• Do not license at all, as set out in alternatives to selling data, including not selling at all.

Where Software Company Licensing Meets Other Industries, Systems and the Seller Path

Software companies share the customer-data question with cybersecurity firms and managed IT providers, and the industry index sets out how each sector's limit differs. The index of data licensing by industry also covers cybersecurity companies licensing data, managed IT service providers licensing data, agencies licensing data and client work and game studios licensing data and assets.

Repository exports follow GitHub repositories: export rights and licenses, issue history travels under Jira data export rights, and copyleft questions sit with open-source license contamination. SaaS switching duties fall under the EU Data Act and AI training data, the full map is AI training data law by jurisdiction, services that train on their own users sit with user data and platform terms, and customer permission sits with customer consent and confidentiality clauses.

1. Check data sale readiness.
2. Build the data inventory for an AI data sale.
3. Apply PII removal and de-identification to what remains.
4. Start choosing a buyer for your data.
5. Work through evaluating an offer for your data.
6. Negotiate the clauses of AI data licensing agreements.

This page is information, not legal advice.

Buyer, Deal, System and Law Pages Named on This Page

Each buyer, deal and rule above has or will have its own page with dates and sources.

Buyers, intermediaries and cases:

• Handshake AI data buyer profile
• HUMXN
• Mercor data buyer profile
• Mercor data breach class actions
• micro1 data buyer profile
• SimpleClosure

Deals, data, systems and law:

• Automattic data licensing
• Connecticut: CTDPA and AI training data
• Cursor and SpaceX AI agreement
• customer support tickets for AI training
• data warehouses as AI training data sources
• export controls and technical data
• FTC rules on AI training data
• HIPAA and AI training data
• issue trackers and project histories
• Oregon: OCPA and AI training data
• Stack Overflow AI licensing deals
• venture-backed startups licensing data

Frequently Asked Questions

Can a SaaS company license data from its product for AI training?

Yes for records the company owns, such as its code, warehouse tables and its own finance, CRM, billing and support records. Customer content in the product only where the customer terms and privacy law allow it.

How much do AI companies pay software companies for data?

No software company price is published by a seller or a court. Mercor states USD 50,000 to 500,000 for code, USD 55,000 to 350,000 for data stacks and USD 20,000 to 480,000 for AI-product sessions, checked 2026-11-09; see what company data is worth to AI companies.

Can a shut-down software startup license its code?

Yes, through whoever still holds authority over its assets. Shutdown intermediaries handle such sales; SimpleClosure states deals in the tens to hundreds of thousands of dollars for unique assets, checked 2026-11-09; see selling data from a shut-down company.

Can software engineers get paid to train AI?

Individual engineers can do paid coding and review tasks for human-data platforms. That is a separate route from licensing the company's records; see AI training jobs and platforms compared.