Yes, for records about running its own platform. A hosting, cloud, colocation or CDN provider can license incident tickets, firewall and WAF change history, DDoS response records and postmortems about its own network. Customer content and customer logs are different: under GDPR Article 28 a provider usually processes them only on the customer's documented instructions, so they stay out unless the customer agrees.
- Split every record into platform (yours) and customer (theirs) before anything else.
- For customer personal data a provider is usually a processor and acts only on the customer's instructions.
- Abuse desk and DDoS records are the provider's own work, but they name complainants and customers who must be removed.
- A single DDoS incident joined to its WAF and routing changes is a strong first sample.
Can a hosting or CDN provider license its incident records to AI?
Yes, when the records are about the provider's own platform. Tickets, change history, WAF and firewall rule changes, routing changes, DDoS response records and postmortems that describe how the provider ran and defended its network are its own records. What a hosting or CDN provider cannot license on its own say is what its customers put on the platform or the logs it keeps for them.
The question to ask of each record is whose activity it describes. This page is general information, not legal advice.
| Record | Whose | Can it be licensed? |
|---|---|---|
| Platform incident tickets and postmortems | The provider's | Yes, after customer names and secrets are removed |
| WAF, firewall, DNS and load-balancer change history | The provider's | Yes, after secrets and customer hostnames are removed |
| DDoS mitigation records for attacks on the platform | The provider's | Yes, with attacker and customer addresses anonymized |
| Abuse desk tickets | The provider's work, about others | Often, with complainants and customers removed |
| Customer websites, files, databases, email | The customer's | No |
| Access and request logs kept for a customer's site | Processed for the customer | Only with the customer's agreement |
Which records belong to the provider and which to its customers?
Records the provider creates to run its network are its own. Records the platform holds because a customer put them there, or because the provider logs traffic to a customer's site on that customer's behalf, are the customer's. Most real tickets mix the two, so the work is cutting the customer out while keeping the incident.
A colocation provider holds the least customer data: power, cooling, access and cross-connect records about its own building. A managed hosting provider holds the most, because its staff work inside customer servers. A CDN sits in between: its edge configuration and attack mitigation are its own, but request logs for a customer's domain describe that customer's visitors.
| Provider type | Mostly its own | Mostly the customer's |
|---|---|---|
| Colocation | Facility incidents, power and cooling events, network edge changes | Equipment inside customer cages |
| Cloud and VPS hosting | Hypervisor, network and storage incidents, platform changes | Virtual machines, disks, customer logs |
| Managed hosting | Platform runbooks, change records, its own monitoring | Work done inside customer servers, customer application logs |
| CDN and DDoS protection | Edge rules, mitigation records, network-wide attack data | Request logs and analytics for a customer's domain |
What does GDPR Article 28 mean for customer logs?
It means a provider usually cannot reuse customer logs for its own purposes. When a customer's site or app collects personal data, the customer is normally the controller and the provider is a processor, which "processes personal data on behalf of the controller" (GDPR Article 4). Article 28 requires the contract to say the processor "processes the personal data only on documented instructions from the controller" (GDPR Article 28).
Two more parts of Article 28 matter. At the end of the service the processor, "at the choice of the controller, deletes or returns all the personal data", so old customer logs may not even be kept. And a processor that decides purposes for itself "shall be considered to be a controller in respect of that processing". Licensing customer logs to train someone else's model is a new purpose, so the provider would need each customer's agreement.
The provider's own security processing is treated differently. GDPR Recital 49 says processing "strictly necessary and proportionate for the purposes of ensuring network and information security" is a legitimate interest of the controller, with "stopping 'denial of service' attacks" as an example (GDPR Recital 49). That covers defending the network. It does not by itself cover licensing the same data later, which is a separate purpose to assess.
What do your own terms and customer contracts say?
Read four documents before choosing records: your terms of service, your data processing agreement, any enterprise contracts with negotiated confidentiality clauses, and your agreements with upstream suppliers such as data center landlords and transit carriers. Many hosting terms promise not to access customer content except to provide the service. That promise covers more than the law does.
- Terms of service and privacy notice: what you told customers you collect and why.
- Data processing agreement: the processing instructions, sub-processor list and deletion terms.
- Enterprise contracts: confidentiality clauses that may cover incident reports and security findings about that customer.
- Upstream agreements: a landlord or carrier may treat its own network details as confidential.
- Security disclosures: incident reports already sent to customers are a guide to what you have said publicly.
Are abuse desk and DDoS response records licensable?
Often yes, because they record the provider's own defense work, but they name people who must be removed. RFC 2142 lists the ABUSE mailbox under customer relations and the NOC mailbox under network operations (RFC 2142), and most providers run an abuse desk that takes complaints, investigates customers and acts. Its tickets show the complaint, the evidence, the decision and the outcome.
DDoS records follow the same pattern. CISA, the FBI and MS-ISAC note that "a DDoS attack involves multiple sources" and sort techniques "into three types: Volumetric, Protocol, and Application." After an attack they recommend a post-incident analysis to "understand its characteristics, vulnerabilities exploited, and attack vectors used" and to "update your incident response plan to incorporate lessons learned from the DDoS attack" (CISA). A provider that did this holds an ordered record of an attack and the response.
- Remove complainant names, emails and attached evidence that identifies people.
- Replace the customer under attack or under investigation with a consistent label.
- Anonymize attacking IP addresses consistently; many belong to compromised home devices.
- Keep the order: detection, rule changes, routing changes, traffic levels, outcome.
Example: one DDoS incident record, end to end
A fictional CDN sees an application-layer flood against one customer's login page. This is the incident record it could offer as one of the three in a sample, after cleaning. Addresses use documentation ranges.
| Time (UTC) | Record | What it shows |
|---|---|---|
| 02:14 | Alert from edge monitoring: request rate to /login at 40 times baseline | Detection |
| 02:16 | Incident ticket INC-4471 opened, on-call engineer paged | Ticket and on-call record |
| 02:21 | Chat thread: traffic from 198.51.100.0/24 and many scattered sources, same user agent | Analysis in order |
| 02:29 | Pull request: new WAF rule rate-limiting /login per source, linked to INC-4471 | The config diff that responds |
| 02:33 | Rule deployed; request rate falls to 3 times baseline | Outcome |
| 03:10 | Second change: challenge page for the user agent; ticket marked mitigated | Second fix and status |
| Day 5 | Postmortem: why the default rate limit did not trigger, two follow-up changes | Lessons learned and follow-ups |
Fictional example written for this page, not a real incident. In the cleaned version the customer is "customer A", the login hostname is replaced, attacking addresses are pseudonymized with a consistent mapping, and payloads are dropped.
Common mistakes when a hosting provider reviews its records
Most mistakes come from treating everything on the platform as the provider's data.
- Including customer request logs because they sit in the provider's log pipeline.
- Forgetting the deletion terms in the data processing agreement and keeping logs that should be gone.
- Leaving customer hostnames and account IDs in WAF rules and change tickets.
- Exporting abuse tickets with the complainant's email and attachments.
- Sending packet captures with payloads. Captures belong in a sample only as part of an incident record, after anonymization.
- Picking an incident still covered by a customer's confidentiality clause or an open investigation.
What do AI data buyers want from a hosting or CDN provider?
AI data buyers have asked Origin for the records of a company running and defending its own infrastructure: incidents joined to the change that fixed them (the ticket, the chat thread, the config diff or pull request, and the outcome), configuration and infrastructure-as-code history with commits and reviews going back years, ordered logs or captures of what happened and what was done, and the runbooks, SOC tickets, on-call records and postmortems of the same period. Hosting and CDN providers were named among the companies that hold them. Records the provider wrote about its own platform qualify; customer content and customer logs do not unless each customer has agreed.
The first step buyers asked for is small: two or three incidents end to end, plus a slice of the configuration history from the same months. For a provider that usually means one DDoS or abuse incident like the example above, one platform outage and the WAF and routing change history around them. Before that, secrets and customer identifiers come out and addresses are anonymized; see anonymizing pcap files and logs and the hub guide. Origin is an independent data partner. Your company decides what is shared and signs any license itself, and a buyer may pay Origin when a deal closes.
Does your company hold records like these? See what your data is worth.
Common questions
Can a web hosting provider license its incident tickets and WAF change history to AI labs?
Yes, for records about running its own platform: incident tickets, firewall and WAF change history, DDoS response records and postmortems. Customer names, hostnames and secrets must be removed first. Customer content and the logs a provider keeps for customers are a separate matter and need the customer's agreement.
Can a hosting company use customer logs to train or license AI models?
Usually not without the customer's agreement. Under GDPR Article 28 a provider processing personal data for a customer acts only on the customer's documented instructions, and a processor that decides its own purposes is treated as a controller for that processing. Terms of service and data processing agreements often add stricter promises.
Is a CDN a data processor under GDPR?
For personal data in a customer's traffic and request logs, a CDN usually acts as a processor for the customer. For its own network security work, such as stopping denial of service attacks, it acts for its own purposes, which GDPR Recital 49 recognizes as a legitimate interest. Each use needs its own check.
Are DDoS attack records personal data?
They can be. Attacking IP addresses often belong to compromised home devices, and records may name the targeted customer. Anonymize addresses with a consistent mapping, replace the customer with a label and drop payloads, and the ordered record of detection, rule changes and outcome stays useful.
Can a colocation provider license its facility incident records?
Yes, records about its own building usually qualify: power and cooling events, access control incidents, network edge changes and the postmortems that followed. Equipment and data inside customer cages are the customer's. Check landlord and carrier agreements for confidentiality terms about the site and its network.
What should a hosting provider remove from abuse desk tickets before licensing them?
The complainant's name, email and attachments, the identity of the customer being investigated, account IDs and hostnames, and any personal data in quoted evidence. Keep the order of complaint, investigation, decision and outcome, with consistent labels so the same customer reads as one party throughout.
Sources and further reading
- GDPR Article 4: definitions (controller and processor)
- GDPR Article 28: processor
- GDPR Recital 49: network and information security as an overriding legitimate interest
- CISA, FBI and MS-ISAC: Understanding and Responding to Distributed Denial-of-Service Attacks (2024)
- RFC 2142: Mailbox Names for Common Services, Roles and Functions
