DataPacts

Cybersecurity Companies Licensing Data to AI

Checked

At a glance

A cybersecurity company can license some of its own alert, case and software records to AI companies, but customer contracts, export rules on dual-use security software and the personal data in its logs decide which records can leave.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

Mercor's Cybersecurity listing names alert-to-resolution records and private security software; Handshake AI, HUMXN, micro1 and Mercor's main program page name none (checked 2026-11-17 and 2026-11-09).

Every figure below is a dated claim from a named source. Security records are one sector in AI training data licensing.

What Cybersecurity Companies Hold That AI Buyers Ask For

Cybersecurity companies hold three kinds of records that a buyer's page describes: the trail from alert to resolution, security software, and the intelligence and rules behind detection. Mercor names the first two; no buyer names the third.

Alerts, Triage Decisions and Case Write-Ups

The record a buyer names for security is the full trail from alert to resolution: alerts from SIEM and endpoint tools, triage decisions, investigation notes and case write-ups.

Mercor's Cybersecurity listing describes "security teams that keep the full trail from alert to resolution", asks for "Several years of cases", names tools such as Splunk, Microsoft Sentinel and CrowdStrike, and lists SOAR and containment logs, detection rules, penetration test reports and "repositories already scrubbed of secrets" as nice to have (read 2026-11-17).

The ticket view of a SOC queue is covered under selling issue trackers and service-desk tickets for AI training.

The listing's scoping questions ask whether case write-ups are kept with the alerts behind them and whether the data is the company's own or its customers', and its "not a fit" list excludes "Logs without case context".

Threat Intelligence, Malware Samples and Detection Rules

No buyer page checked names threat intelligence feeds or malware samples as a record type, so a company that holds them cannot read demand from a buyer's page. Mercor lists detection rules and penetration test reports only as nice to have, and Handshake AI, HUMXN and micro1 name no security record.

The rights to a feed a company ingests, to samples received from customers or sharing groups, and to detection rules written on a customer's engagement are each a separate question that this page does not answer. No cybersecurity data deal is in the deal tracker as of 2026-11-17.

Buyer and Marketplace Statements That Name Security Records

One program names security records on its own page, checked 2026-11-17, and four other statements checked name none. The table compares what each states for security as of 2026-11-17, and each value is the party's own statement.

Program Role What it names for security, or does not Source
Handshake AI B 11 industries, none security (checked 2026-11-09) joinhandshake.com/ai/data-partnerships
HUMXN M 24 industries, none security; "Software & SaaS" lists code, issues, incidents, support tickets, product analytics (re-read 2026-11-17) humxn.io/sell-company-data
Mercor Cybersecurity Company listing B alert-to-resolution records, investigation notes, case write-ups, private security software; 20 to 2,000 employees; USD 20,000 to USD 750,000 depending on case volume and rights (claim); excludes "Logs without case context" and "Customer data the company can't share"; customer data only with customer agreement api.data.mercor.com listings /cybersecurity
Mercor Data program page B 8 target industries, none security (re-read 2026-11-17) mercor.com/data
micro1 B six functional areas, none security (checked 2026-11-09) micro1.ai/data-partnerships

B = buyer or licensee program; M = marketplace. "Not stated" does not mean allowed.

The Mercor deal range of USD 20,000 to USD 750,000 "depending on case volume and rights" is the buyer's own claim read 2026-11-17, not a market price, and the listing's referral terms are not stated here.

"Not stated" on a program page does not mean the program is closed to security records. Each program's payouts and terms are set side by side in payouts and program terms compared.

Customer Contracts and Third-Party Terms Come Before a License

A security company holds much of its record trail for someone else, so the contract that gave it access decides what it may license before any buyer's wish list does. Mercor's listing requires "The right to license them; customer data only with customer agreement" (read 2026-11-17).

Incident Reports and Customer Environment Data

Investigation notes, incident reports and penetration test reports describe a customer's systems, staff and weaknesses, so the customer's agreement is the first condition. Mercor excludes "Customer data the company can't share", and a penetration test report is a client deliverable licensed only if the engagement terms allow it.

The managed-provider view, with client service agreements and processor duties, sits on the managed IT provider page named below.

Feeds, Sharing Groups and Third-Party Intelligence

Intelligence a company receives from a commercial feed, a customer or a sharing group is held under that source's terms, and this page states none of those terms.

Dual-Use Software and Export Rules for Security Records

Security software and technology for intrusion tools sit on the Commerce Control List, so a company decides which files are controlled before it describes a set to a buyer or lets a foreign annotator see it.

What 15 CFR 740.22 Lists as Cybersecurity Items

15 CFR 740.22(b)(1) defines 'cybersecurity items' as a list of ECCNs, among them 4A005, 4D004 and 4E001.c, with 4D001.a and 4E001.a for those items and 5A001.j with its related entries (eCFR text as of 2026-11-13, read 2026-11-17).

License Exception ACE in 740.22(a) authorizes export, reexport and transfer of those items, including deemed exports, subject to 740.22(c).

ECCN 4D004 covers "software" specially designed or modified for the generation, command and control, or delivery of "intrusion software"; ACE applies except to Country Group E:1 or E:2, and STA may not be used.

ECCN 4E001.c covers "technology" for the "development" of "intrusion software", and Note 1 to 4E001 says that 4E001.a and .c do not apply to "vulnerability disclosure" or "cyber incident response". 15 CFR 772.1 defines "intrusion software" as "software" specially designed or modified to avoid detection by monitoring tools or defeat protective countermeasures and to extract data or modify the standard execution path of a program or process.

ACE does not authorize an item to Country Group E:1 or E:2 (740.22(c)(1)), to a government end user in Country Group D:1 to D:5 with stated exceptions ((c)(2)), or where the exporter knows or has reason to know of use to affect the confidentiality, integrity or availability of information systems without authorization by the owner, operator or administrator ((c)(5)).

Malware Samples, Exploit Code and a Foreign Annotator

Releasing controlled technology or source code to a foreign person in the United States is a deemed export under 15 CFR 734.13, so a security file can be exported to an annotator without leaving the country.

Under 734.13(b), the destination is the person's most recent country of citizenship or permanent residency, and 740.22(a) authorizes deemed exports of cybersecurity items within its restrictions. Whether a malware sample, exploit proof of concept or detection tool falls within 4D004 or 4E001.c is a classification question for the company, not answered here.

micro1's master terms, section 5.7, exclude "viruses, malware or other harmful or destructive code" and export-controlled or ITAR/EAR-restricted information unless the Order Form allows it (terms updated 2026-11-12, re-read 2026-11-17). The general rules sit with export controls and licensing technical data to AI buyers.

Personal Data in Logs and Alerts: Privacy Rules Before a License

Logs and alerts name people through addresses, usernames and device identifiers, and each rule below attaches to the person identified. The table compares two rules as of 2026-11-17.

Jurisdiction Rule What it says for logs In force Source
European Union Regulation (EU) 2016/679 (GDPR) Art. 4(1), recitals 30 and 49 personal data includes identification by an online identifier; recital 30 names IP addresses; recital 49 on security processing applies from 2018-05-25 eur-lex.europa.eu
United States 28 CFR 202.212(c)(4), 202.206 and 202.205(f) (DOJ Bulk Data Rule) name plus email plus IP address is a covered personal identifier; anonymized data stays covered; threshold more than 100,000 US persons effective 2025-04-08 ecfr.gov

Other privacy statutes, state breach-notification duties and sector rules are not yet summarized on this page.

IP Addresses, Usernames and Credentials in Logs

An IP address, an account name or a device identifier in a log can make a record personal data, so a security company licenses logs only under the privacy law of the people they name.

Article 4(1) of Regulation (EU) 2016/679 counts a person identifiable by "an online identifier", and recital 30 names "internet protocol addresses, cookie identifiers" among them; the rule applies from 2018-05-25, and the legal basis is set out under GDPR and AI training data.

Recital 49 treats processing strictly necessary and proportionate for network and information security, including by providers of security technologies and services, as a legitimate interest of the controller.

Recital 49 concerns the security purpose; recital 50 allows other purposes only where compatible with the initial ones, and whether an AI license is compatible is a question for the company's counsel.

Under 28 CFR 202.212(c)(4), a name linked to an email address and an IP address is a covered personal identifier, and under 202.206 bulk US sensitive personal data stays covered when anonymized, pseudonymized, de-identified or encrypted, with a threshold of more than 100,000 US persons in 202.205(f).

Mercor lists repositories already scrubbed of secrets only as nice to have; the prohibitions are set out under the DOJ bulk sensitive data rule.

Analysts and Employee Records

Case notes and ticket histories name the analysts who wrote them, so employee data rules apply next to the company's rights.

Benefits, Risks, Downsides and Alternatives for a Cybersecurity Company

Licensing can earn income from case records and security software a company keeps anyway, but customer, export and privacy limits shrink what can go, and not licensing is a valid outcome.

Benefits

  • A buyer names alert-to-resolution records and private security software.
  • A company's own records need no customer consent where they hold no customer data.

Risks

  • Customer data leaving without the customer's agreement.
  • Export of controlled software or technology, including to a foreign annotator.
  • Personal data in logs, and secrets left in repositories.

Downsides

  • Screening work to separate own records from customer records.
  • Mercor excludes logs without case context and sets a band of 20 to 2,000 employees.
  • micro1 excludes malware and export-controlled information unless the Order Form allows it.

Alternatives

Where Cybersecurity Licensing Meets the Seller Path, the Law and Other Industries

Security companies share the customer-data question with managed IT providers and software companies, and each industry page sets out its own limits. See data licensing by industry, managed IT service providers licensing data, software companies licensing code and product data and consulting and advisory firms licensing client deliverables, since penetration test reports are client deliverables.

Security software follows the rules for selling or licensing source code to AI companies, and case write-ups follow the rules for selling documents for AI training. Researcher work on AI models is a different page, red-teaming data and adversarial testing work.

The law is mapped in AI training data law by jurisdiction; log records fall under privacy law and AI training data, analyst records under employee data and AI training, and detection rules and customer findings raise trade secrets and confidential information in AI data licensing questions.

  1. Check data sale readiness for the records.
  2. Run a rights audit of customer-contract records.
  3. Review customer consent and confidentiality clauses.
  4. Remove personal data through de-identification before licensing data.
  5. Plan security during an AI data sale.
  6. Negotiate the clauses of AI data licensing agreements.

This page is information, not legal advice.

Buyer, Marketplace and Law Pages Named on This Page

Each buyer and marketplace above has its own page.

Frequently Asked Questions

Can a security company license threat intelligence to AI companies?

No buyer page checked names threat intelligence as a record type, so the company needs the rights to each feed and sample first, and those terms are not summarized here.

How much do AI companies pay security companies for data?

Only Mercor's Cybersecurity listing states a range, USD 20,000 to USD 750,000 depending on case volume and rights, its own claim read 2026-11-17; no other program checked publishes a security price.

For context, see what company data is worth to AI companies.

Can a security firm license malware samples or exploit code?

Not by default: micro1 excludes malware unless its Order Form allows it, and ECCN 4D004 and 4E001.c can control intrusion-software code and development technology, including on release to a foreign person.

Byline: DataPacts Editorial Team, see how we source, date and check every figure.