One control layer for every path to AI.
Employees reaching AI applications, enterprise applications calling models through APIs, internal AI applications and AI agents all take different routes to the same models. DSX Guard governs every one of them from one place, across multiple AI and model providers, with one policy set and one audit trail. For the category this sits in, see what enterprise AI governance means in practice.
Protect
Mask or block sensitive values, inspect attached files, and detect threats before anything leaves your perimeter.
Govern
Apply organisational policy to every request, record what happened — including which AI tools employees actually use — and stream the evidence into the tools your security team already uses.
Connect
Give AI governed access to the systems that hold the answer — through managed agents and connectors, never through pasted credentials.
One security and governance layer across human-to-AI and application-to-AI interactions.
Protect AI directly in the browser.
Employees keep using the AI tools they already know. DSX Guard inspects every interaction before it reaches the model — no new application to learn, no workflow to change.
Deploy once. Protect every managed device.
Install the DSX Guard agent on the devices your organisation chooses to protect. From then on, the Guard admin console centrally deploys and manages the browser extension across Chrome and Microsoft Edge — without requiring employees to install, configure or maintain anything.
Devices with the DSX Guard agent are automatically brought under protection and managed from the central console.
Employees cannot disable or remove the extension through normal browser controls. Organisational policies keep protection active across managed devices.
Administrators are alerted if protection is interrupted, removed outside policy or otherwise falls out of compliance.
Extension versions are updated centrally, keeping every protected device on the approved release without employee action.
Centrally deployed. Policy protected. Always on.
Protect the data. Preserve the workflow.
Sensitive values are replaced before the request leaves the browser and restored when the answer comes back. The employee never sees the difference. Why blocking public AI fails sets out the reasoning behind this model.
The employee gets the answer. The AI provider never gets the sensitive data.
Masking that survives arithmetic.
Substitution alone breaks the moment a question operates on the values — a model cannot total two amounts it never saw. Guard handles this by letting the model express the operation over the placeholders. The arithmetic is then evaluated against the real values inside your tenant, and the result is placed into the answer.
The employee reads a complete, correct answer. The provider never held a single underlying value.
Your data. Your definitions. Your policies.
Sensitivity does not mean the same thing in a bank, a hospital and a software company. Define it for your organisation, then decide what happens when it appears.
Personal data
Names, national identifiers, email addresses, phone numbers.
Financial data
IBANs, account numbers, card and payment information.
Business-critical data
Customer IDs, contract numbers, pricing, internal codes.
Technical secrets
API keys, credentials, tokens, source-code patterns.
Custom sensitive data
Organisation-specific patterns, dictionaries and rules.
Stop unsafe AI interactions before they happen.
Prompts and files are analysed on every request — not sampled after the fact, and not left to the model provider to police.
Detect
Every prompt and attached file is analysed against known attack patterns and your own content rules.
Evaluate policy
Findings are scored against the rules your organisation set — by user, group, application and data category.
Allow · Block · Log
The request proceeds, is stopped, or is recorded for review — and the event reaches your SIEM either way.
Bring enterprise data into the AI tools employees already use.
DSX Guard adds governed enterprise agents directly into the assistant the employee already has open. The agent reaches your systems under enterprise permissions — the employee just asks the question.
Employees don't need another AI application. Enterprise capability comes to the AI they already use.
Register DSX IQ in Guard and your document archive becomes one of those agents. An employee asks a question in ChatGPT; Guard masks what policy requires, routes the question to IQ, and returns an answer drawn from your corpus with the document, page and passage attached. Every exchange is inspected and logged like any other.
Connect AI securely to enterprise data.
Agents reach real systems through managed connectors — never through credentials pasted into a prompt.
Give agents the data they need — under enterprise-defined permissions and policy, with every access recorded.
Secure application-to-AI traffic.
DSX Guard is also the central gateway for every application in your estate that calls an LLM — one endpoint instead of a provider SDK in each service.
One integration
Applications target a single endpoint instead of a different provider SDK in every service.
Swap models without redeploying
Routing is configuration, not code — move traffic between providers centrally.
Keys never leave the platform
Provider credentials stay in DSX Guard instead of being copied into each application.
Both paths. One rule set.
Everything above converges here: whether a person or a service is talking to the model, it meets the same evaluation and leaves the same trail.
Human-to-AI and app-to-AI. Governed together.
One console for the whole control layer.
Policy, infrastructure, agents and evidence in one place — not spread across six separate tools.
For organisations where data matters.
Protect customer, account and transaction information.
Protect policyholder and claims data.
Protect confidential and privileged information.
Protect sensitive health and operational information.
Protect source code, credentials and intellectual property.
Questions about AI security and governance.
Does this block employees from using AI?
Only where your policy says so. Blocking, warning, masking and allowing are all outcomes, evaluated per user, per assistant and per data class. Masking is the default posture because it is the one employees do not route around.
Does the AI provider still see the prompt?
It sees the structure of the task with placeholders where sensitive values were, which is what makes the answer useful. The values themselves never leave, and they are restored only inside your tenant.
Does masking break answers that need the real numbers?
No. Where a question operates on masked values — totalling two amounts, comparing two dates — the model returns the operation over the placeholders and Guard evaluates it against the real values inside your tenant, then writes the result into the answer.
Is this the same as our DLP?
It is the AI-path extension of it. Traditional DLP was built for email, endpoints and file transfer; browser AI prompts and application-to-model API calls are channels it does not see. Guard covers those and streams its events into the DLP and SIEM tooling you already operate.
What about applications and agents, not just people?
Applications call models through a gateway that holds the provider keys centrally instead of each service carrying its own, under the same policy set and one audit trail. Agents connecting to enterprise systems run under the same controls, with human review where policy requires it.
Which assistants and providers are covered?
Browser protection covers the major AI assistants employees already use, and the gateway routes application traffic across providers with central key management, so provider choice stays a configuration decision rather than a rebuild.
What does an auditor actually get?
Who asked what, which assistant or application it went to, which data classes were masked, which policy applied and what the outcome was — recorded in a form that can be read without an engineer and streamed to your SIEM.
How is it priced?
A shared platform base plus a per-protected-user rate that steps down with volume. Applications calling models through the gateway are priced per application by traffic volume, quoted with your numbers. Every capability is included — there are no feature tiers.