Active by default
Every company starts with protection enabled. There is no switch that someone could forget to turn on.
Active anonymization
Sensitive information stays in your organization; the task reaches the model
Companies do not postpone adopting AI for lack of interest, but because they cannot allow a contract, a medical record or a client's name to end up in the hands of an external provider. NiuCore addresses that risk before the message leaves the platform, without affecting the quality of the responses.
The risk
AI is already being used in your organization; the question is with what information
When the company does not provide a corporate tool, each employee finds their own solution: personal accounts, documents pasted into a chat, client emails summarized outside any control. This is what is known as shadow AI. Prohibition is not effective and training is not enough, because one oversight is all it takes. Reliable protection is the kind that does not depend on individual memory or judgment.
- Client and patient data in personal tools
- No record of what information was shared
- Policies whose compliance depends on each employee
What your team writes and what the model receives
The employee's request
Draft an email to Sensitive data: Ana Pérez (Sensitive data: ana.perez@clinicaaurora.co, Sensitive data: +57 310 555 0142), manager at Sensitive data: Clínica Aurora S.A.S. (tax ID Sensitive data: 900.482.117-3), about the renewal of contract Sensitive data: CT-482 for the Sensitive data: Horizonte project. Say that Sensitive data: Carlos Ruiz, ID number Sensitive data: 52.418.993, will be the new account lead, that payment will be charged to card Sensitive data: 4111 1111 1111 1111, and copy Sensitive data: carlos.ruiz@clinicaaurora.co.
Text sent to the model (ChatGPT, Gemini and others)
Draft an email to Placeholder: ⟦Person#7F3A2C⟧Automatic detection (Placeholder: ⟦Email#A1B2C3⟧Format, Placeholder: ⟦Phone#5D20E7⟧Format), manager at Placeholder: ⟦Company#C41F09⟧Automatic detection (tax ID Placeholder: ⟦Document#9A3E61⟧Format), about the renewal of contract Placeholder: ⟦Contract#4E91D0⟧Dictionary for the Placeholder: ⟦Project#0F9C2B⟧Automatic detection project. Say that Placeholder: ⟦Person#2B8D4F⟧Automatic detection, ID number Placeholder: ⟦Document#B81E44⟧Format, will be the new account lead, that payment will be charged to card Placeholder: ⟦Card#E07A13⟧Format, and copy Placeholder: ⟦Email#6C9B21⟧Format.
Response the employee receives
Subject: renewal of contract Sensitive data: CT-482 Dear Sensitive data: Ana: we are pleased to confirm the renewal of contract Sensitive data: CT-482 for the Sensitive data: Horizonte project with Sensitive data: Clínica Aurora S.A.S.. From today, Sensitive data: Carlos Ruiz will lead your account; you can reach him at Sensitive data: carlos.ruiz@clinicaaurora.co. Payment will be charged to the registered method…
The third column is what your team receives. The second is everything the model knows.
Response quality
The same response quality, with your information protected
Protecting data should not mean giving up the best models or settling for inferior answers. Active anonymization preserves the structure and context of each task—who is involved, what the document is, how the parties relate—so the model reasons in the same way and the quality of the response is not affected. The employee receives the result with the real data restored.
- Context and relationships between data points preserved
- Consistent placeholders throughout the conversation
- Access to frontier models without exposing sensitive information
Layer 1 · Formats
Data with a recognizable format is protected automatically
Email addresses, phone numbers, national IDs and tax IDs with check digits (such as RUT and NIT), card numbers and bank accounts. NiuCore identifies them by their format and replaces them with no prior configuration. The company decides which data types it keeps active.
- No initial setup
- Format validation, not simple matches
- Each data type can be enabled or disabled
Layer 2 · Dictionary
The information only your organization recognizes as sensitive
No generic detector knows that "CT-482" is one of your company's contracts or that "Aurora" is a confidential project. In the dictionary you define exact terms or patterns—for example, any reference starting with "CT-" followed by numbers—and they are protected across the organization from that moment on.
- Exact terms or patterns
- Centralized administration for the whole company
- Changes applied immediately
Layer 3 · Automatic detection
NiuCore identifies the vocabulary specific to your business
From the company's content, NiuCore proposes the terms worth protecting: people, organizations, products, brands and projects. Each proposal protects from the moment it is detected, on a provisional basis, and a designated person in the company confirms it or discards it if it is a false positive.
- Protection from detection, without waiting for approval
- Human review to refine the result
- Initial dictionary built from the company's website, if provided at registration
Protection that does not depend on individual judgment
Designed so that the protected path is the only one available.
Beyond the employee's reach
Anonymization is governed at company level. Turning it off requires a request that NiuCore reviews.
Documents included
Libraries are indexed with data already replaced. The information the model consults is already protected.
Recommendations included
If NiuCore cannot anonymize the information needed to recommend a next step, it does not generate the recommendation. Protection takes priority.
Scope and limits
What your organization should know before subscribing.
Scope
Employees' messages and conversation history, library documents, attachments, images and scanned documents, information delivered by connectors, and next best action recommendations.
Reversible by design
Placeholders are restored so the employee receives the response with the real data. It is not an irreversible removal of identity.
It is not infallible
No detector recognizes every piece of sensitive data. That is why there are three layers and human review, and why the dictionary is managed by the company.
It applies to what is processed in NiuCore
It does not intercept other applications. Its value lies in offering the organization a protected path that people prefer to use.
Protection from day one
Company registration
Anonymization is already active and, if the website is provided, NiuCore builds an initial dictionary from it.
Review of proposed terms
A designated person confirms those that apply and discards those that do not.
Adding your own patterns
Contract numbers, case codes, internal identifiers.
Onboarding the teams
Employees work as usual; protection operates transparently.
Frequently asked questions
Is this the same as the "incognito mode" in the API documentation?
Yes. The technical documentation calls it incognito mode; the product and this site call it active anonymization.
Can the model infer the original data from the context?
The model receives placeholders, not the data. The context of the task is preserved—the existence of a person, a contract, a date—but not its real value. If a text describes someone in such detail that they are identifiable without being named, no replacement system solves that completely; the dictionary allows those terms to be protected as well.
What happens if NiuCore flags a term as sensitive when it is not?
It is discarded during review and is no longer applied.
Where is the link between the placeholder and the real data kept?
In NiuCore's infrastructure; it is not sent to the model provider. More information in Trust.
Does it work with images and scanned documents?
Yes. Anonymization is also applied to the content of images and scanned documents before it is sent to the model.
Can it be turned off for a specific area?
Anonymization is governed at company level. What is configured per area is its members, models, connectors, libraries, rules and skills.
Active protection in every license
Active anonymization is part of every plan, at no additional cost.