Data Protector

Catch sensitive content before it leaves your environment.

What Is Data Protector?

Your team shares information with AI tools every day. Data Protector makes sure sensitive content is caught before it leaves your environment. It reviews every trace for the categories you care about and acts according to its role.

This is your primary safeguard for GDPR, data sovereignty, and privacy obligations.

Data Protector is a paid add-on. If it is not enabled for your organisation, contact us to turn it on.

What It Catches

You choose which categories Data Protector looks for and prevents being sent.

Credentials

Secrets used to access systems: passwords, API keys, access tokens, private keys, and connection strings.

Personal data

Information that identifies a person. Aligned with GDPR, and split into three sub-categories you can enable independently.

Ordinary personal data. Everyday details that can identify someone: name, contact details, date of birth, government identifiers (such as a CPR, SSN, or passport number), and device or location data tied to a named person. This is personal data in the GDPR Article 4 sense that is not a special category.

Special categories of personal data. Data given extra protection under GDPR Article 9 — often called sensitive personal data. That list is health, biometric data processed for the purpose of uniquely identifying a natural person, and genetic data, racial or ethnic origin, political opinions, religious or philosophical beliefs, trade-union membership, and data concerning sex life or sexual orientation.

Other confidential data. Confidential information about a person that is not an Article 9 special category, such as salary, HR or disciplinary records, and family circumstances.

Financial data

Payment and banking details: card numbers, IBANs and bank account numbers, and similar payment identifiers.

Where the categories come from

The categories are Velatir’s. We defined this taxonomy for AI usage. It is not imported from a third-party DLP catalogue, and Data Protector does not read classification labels from tools such as Microsoft Purview or Google Cloud DLP, or from labels you already maintain elsewhere.

• Personal data is aligned with GDPR. Ordinary personal data follows Article 4 (information that identifies a person). Special categories follow Article 9 (the types of data given extra protection).
• Credentials and financial data are operational categories we defined for secrets and payment information, following common DLP practice rather than a single regulation.
• Intellectual property is available so you can attach your own instructions — for example a product codename. Velatir does not ship a built-in detector for it.

If you already classify data in another system, encode the extra rules you care about as instructions. Instructions sit alongside these categories; they do not replace them.

The taxonomy is GDPR-oriented so you can turn the right protections on. It is not a legal determination that a given prompt is or is not personal data under your policies.

How it looks at content

Data Protector looks for both known formats (for example a card number or an API key) and meaning in the text (for example health information about a named person). Special-category and other confidential findings are treated as sensitive when they are tied to an identifiable person. Asking how to handle health data as a policy question is not treated the same as sharing a named person’s medical details.

Detection is not limited to English. Field labels, names, addresses, and identifiers are recognised in other languages too.

Choose What It Assesses

Capabilities decide where the Data Protector looks. Choose what the agent is able to assess.

Capability	What it assesses
Input assessment	The agent assesses text entered into AI services.	Always on
File assessment	The agent also assesses the contents of uploaded files.	Beta
Clipboard assessment	The agent assesses content pasted from the clipboard.	Beta

With Clipboard assessment on, you can turn on Redaction for a category, so sensitive content is stripped from text users paste into AI tools — analysed on the device, before it is sent.

Configure Each Category

For every category you turn on, you can set how seriously to treat it.

Setting	What it does
Enabled	Turn the category on or off.
Criticality	Mark findings in the category as high criticality. High-criticality blocks also notify your connected channels.
Redaction	On categories that support it, strip matching content from pasted text on the device. Requires Clipboard assessment and Enforcer. See Redaction.
Sub-categories	For personal data, choose ordinary, special-category, and other confidential independently.

Choose a Role

Role	Behaviour
Observer	Reviews traces and flags findings for review, without notifying anyone or blocking.
Enforcer	Blocks sensitive content before it reaches the AI service. High-criticality findings also notify your channels.

New Data Protectors start as Enforcer. Move to Observer first if you want to learn what it catches before you enforce.

False-Positive Challenges

Turn this on per Data Protector when you are ready to block with a safety valve. If the agent blocks something it should not have, the person can challenge the block, explain why, and keep working. No dead end, no support ticket.

Challenged assessments are marked on Assessments, and a False positive tab shows the reasoning they gave. Use that feedback to tune categories and instructions.

An optional grace period keeps a reporting user unblocked for five minutes, so Data Protector does not become a bottleneck while the challenge is on record. The grace period is a separate toggle.

Tune It With Instructions

Categories cover the common cases. Instructions let you handle the specific ones, such as always blocking a project codename or allowing a particular scenario your team has reviewed. Instructions are grouped by category on the agent.

When to Use Enforcer

Promote Data Protector to Enforcer when your organisation handles regulated personal data, operates in healthcare or financial services, or is subject to GDPR enforcement. The cost of a leak in these settings outweighs the occasional blocked trace. Turn on false-positive challenges so a wrong block has recourse. For teams handling mostly non-sensitive work, Observer gives you visibility without interruption.

---

Redaction

Strip sensitive content from pasted text, on the device.

Instructions

Add your own rules for specific scenarios.

Configuring agents

Defaults, extra Data Protectors, and workspace assignment.