Most AI announcements in the domain industry focus on what an agent can do. They detail how many tasks it covers and how many tools it connects to. An infrastructure provider that resells domains to its own customers needs to know how much authority the AI has, and when a person steps in to make the call.
Openprovider AI runs in four places:
- Trust and abuse scoring
- Identity checks at signup
- Support automation
- A domain management co-pilot.
Openprovider’s in-house AI team builds and runs each one around the same question: is the system confident enough to act, or should a person decide?
Our blog on solving domain management with AI covered the general benefits of AI. This article shows how our own systems work, starting with the one rule they all follow.
One rule keeps a person in charge of every uncertain decision on your domains
Every Openprovider AI system follows the same rule: confidence decides who acts.
Each system gives every decision a confidence value, a score for how sure it is about its own answer. With high confidence, the system acts on its own, and with lower confidence, it sends the case to a person. When it has too little information to judge, it takes no action.
Abuse scoring is the first place we apply the rule, because a wrong decision there affects your customers’ domains.
Risk scores with confidence values protect your customers from wrongful suspensions
Openprovider AI gives every domain on our platform a risk score from 0 to 100, plus a confidence value and a risk band, a label that groups scores into levels of risk. Our AI team built the score from around ten independent analyzers that run in parallel. Each analyzer checks a different kind of evidence, such as how the domain name is built or whether it imitates a known brand through a small spelling change, a tactic called typosquatting.
The system resolves obvious cases in milliseconds on a fast path and runs ambiguous ones through every analyzer before it acts or passes the case to our abuse team. Around the score, a classifier sends each abuse report to the right queue, bulk classification sorts large takedown requests, and anomaly detection flags sudden changes in abuse volume. Our article on how Openprovider handles domain abuse explains the policy and suspension process.
For a reseller, this means we stop real abuse faster and suspend fewer legitimate customers by mistake, so your team handles fewer escalations from clients. Abuse handling also affects your reputation: Interisle’s Phishing Landscape 2026 research found that attackers registered about three quarters of phishing domains in recent years specifically for phishing.
We keep the weights, thresholds and specific signals behind these scores private, because publishing them would show abusers how to avoid detection. Our identity checks start even earlier, before an account can register its first domain.
Real-time identity checks keep bad actors off your platform from the first signup
Openprovider AI checks every new account in real time with a set of small models, each built for one task:
- Validating documents
- Authenticating users
- Matching faces
- Detecting AI-generated documents.
Together, these models run KYC and KYB checks (know your customer and know your business) during signup. People on our team review only the edge cases, so legitimate businesses sign up quickly.
Once an account is live, the same confidence rule helps Isabella answer its support tickets faster.
Faster support answers, with a person checking every uncertain reply
Isabella, our support automation tool, uses classical machine learning to read each ticket, work out what the customer needs and find the right knowledge base article. A large language model then writes the reply.
Isabella sends high-confidence replies automatically, saves medium-confidence replies as drafts for a support agent to review, and leaves low-confidence tickets for a person to handle from the start. It works in four languages and learns from every reply an agent edits or approves. Isabella is live in our support team and rolling out as chat in the Reseller Control Panel. Customers get faster first responses, and Isabella resolves more tickets without a person getting involved. Our article on AI and automation in hosting covers the wider industry.
Isabella answers questions, while the co-pilot takes actions on your domains, so its permissions matter even more.
The co-pilot automates domain work with the permissions your team already has
The Openprovider co-pilot lets your team manage domains by typing requests in plain English. Staff without API skills can ask it to find a domain, calculate costs, explain a setting, open the right screen or carry out a task.
Some platforms give AI agents admin-level access. Our AI team designed the co-pilot to act only as the logged-in user, with that user’s own access token. It has no admin identity and no direct access to our database.
The co-pilot can immediately run read-only tasks, such as searches and calculations. It can only perform write actions from a fixed list, so it cannot create new operations. For most write actions, it shows a preview card with the exact action and settings and waits for the user to approve, and deletes and bulk changes need a separate hard confirmation. Every action goes through the Openprovider API and its standard checks.
The co-pilot uses the role and permissions you already gave each team member, so you can delegate domain work to more of your team without giving anyone extra access. Use these five questions to check how much authority any AI has over your customers’ domains.
| Question to ask any AI that touches your domains | Openprovider’s answer |
|---|---|
| Whose identity does it act with? | The logged-in user’s, using that user’s own token. It has no admin identity. |
| Can it invent new operations? | No. It can only perform write actions from a fixed list. |
| Which actions need approval? | Most write actions, after a preview card shows the exact action and settings. |
| Which actions need a hard confirmation? | Deletes and bulk changes. |
| Can it see more than the user can? | No. It has no direct database access and sees only what the user’s account allows. |
We apply the same care to billing, where AI checks that Members pay the right price.
Continuous billing checks keep cost-price Membership pricing accurate
Registries start and end promotions all the time, and each change affects what Openprovider pays and what Members pay. Openprovider AI compares the two prices continuously and flags any gap as soon as it appears. It checks both directions: that Members get registry discounts while they apply, and that Member prices update when a promotion ends. This keeps cost-price pricing accurate on eligible TLDs, so you can plan your margins on correct numbers.
Clear AI rules and human sign-off protect every change to your platform
The Openprovider AI team has built finance automations, automated QA testing and a pipeline that turns a Jira ticket into a pull request, a proposed code change ready for review. These tools have saved our organization tens of hours of manual work so far, with hundreds more expected once planned projects ship. The team is also testing a model-agnostic “software factory” that can run on different AI models, so Openprovider does not depend on a single AI vendor.
An AI ambassador program scores teams on the AI-powered processes they ship, our Security team owns the list of approved AI tools, and a person signs off on any significant or customer-facing change. The team is already building the next tools for the Reseller Control Panel.
AI-driven onboarding and an AI website builder are on the roadmap for your Reseller Control Panel
Two new tools are on the roadmap for the Reseller Control Panel: an AI-driven onboarding flow and an AI website builder. Neither has shipped yet, and we will share details closer to release.
Registrar-scale data gives Openprovider AI an edge that competitors cannot license
Any company can license the same AI models we use. Openprovider has built the data those models learn from over more than 22 years: registration, abuse and reconciliation records at registrar scale. This data makes our models accurate, and a culture of testing and shipping keeps improving them.
If you are deciding whether to trust a registrar’s AI with your customers’ domains, start with the questions below.
Questions to ask before you trust Openprovider AI with your domains
How does Openprovider use AI?
Openprovider uses AI in four places: trust and abuse scoring, identity checks, support automation and a domain management co-pilot. It also uses AI to check billing. Each system acts alone only when it is confident.
Is it safe to let an AI agent manage domains?
It depends on the agent’s authority. The Openprovider co-pilot acts only as the logged-in user, uses a fixed list of write actions and asks for approval before most changes.
What does human in the loop mean for Openprovider AI?
A person reviews every case the AI is not confident about, and the co-pilot waits for the user’s approval before most changes.
Who builds Openprovider AI?
Openprovider’s in-house AI team builds and runs every system in this article.
