EnterpriseReal-Time AI Data & Intent Security
Data = what AI receives · Intent = what AI may do
Safeguard data. Govern AI intent. Give every employee a governed path to use ChatGPT, Copilot, coding agents and AI tools of their choice — without exposing regulated data or letting an agent take an action it was never authorized to take.
Make enterprise AI safe, private, and governable.
One management console, every employee device
A lightweight endpoint agent inspects AI use on every device; one console sets the policy and keeps the audit trail.
Management console
Cloud or on-premises- Deploy endpoints
- Create and distribute rules
- Collect events and alerts
- Apply employee/device policy
- Audit who, what, and when
Endpoint agent
Windows · macOS · cloud-deployed agent environmentsTwo fundamental enterprise AI challenges
AI adoption is spreading across every workflow. Guardrail defends at the exact moment an employee interacts with ChatGPT or an AI agent acts — closing the security blind spot that opens up in between.
Enable AI use without exposing sensitive data
- Encourage employees to use AI productively
- Share enough business context to produce useful results
- Prevent exposure of regulated, proprietary, or confidential data
Natural language can become a harmful action
- Detect executable intent in prompts, code, and agent messages
- Intercept unauthorized commands, file access, transactions, and subscriptions
- Allow, warn, require approval, or block — in real time
Sandboxing or containerization can isolate an AI agent — but cannot prevent it from carrying out an unauthorized action it has been instructed, or manipulated, into taking.
AI intent governance: 10 categories of high-risk intent, examined everywhere
Guardrail inspects the meaning behind a prompt, document, or command — not just its keywords or a known signature — and evaluates it against policy before an employee or AI agent can act.
The real-time AI interaction pipeline
Deep inspection detects both sensitive data and executable intent — what traditional cybersecurity tools, built to inspect signatures and network traffic rather than meaning, cannot.
Input sources
- Prompts & chat
- File uploads
- Clipboard & screen
- Local files & folders
- Agent tool calls
- MCP & agent-to-agent
iDox.ai Guardrail™
- Sensitive-data detection
- Context & intent analysis
- Policy evaluation
- Risk assessment
Policy action
Destinations
- Chat & assistant AI
- Productivity copilots
- Coding agents
- Custom & open-source models
- Other agents & MCP tools
Detection, transformation, and policy decisions happen at the point of AI use — before content reaches its destination model or agent.
Built for meaning, not just signatures
Traditional cybersecurity tools inspect signatures and network traffic. Guardrail inspects what content means and what an agent is being instructed to do.
Adaptive, self-learning detection
Built-in intelligence continuously improves detection of sensitive data and emerging threats.
Contextual sensitive-data detection
Recognizes regulated and enterprise-specific information using language, surrounding context, entity models, and configurable rules.
Privacy-preserving transformation
Masks, redacts, pseudonymizes, or anonymizes sensitive values before disclosure, while retaining useful document structure and context.
AI-agent action governance
Evaluates file access, tool calls, generated commands, and external communications against policy — high-risk actions can require human approval.
Natural-language threat analysis
Identifies prompt injection, harmful instructions, and suspicious generated commands that may not match a known signature.
Central policy and audit
Policy distribution, event collection, risk visibility, alerts, and Who-What-When records across every protected endpoint.
Guardrail combines a lightweight endpoint agent with centralized management. Detection and intervention happen close to the employee or AI-agent workflow; the console distributes policy, collects events, and governs the entire deployment — so every employee stays aligned to one AI usage policy.
Detect & sanitize, in real time
Names and amounts are detected and replaced with policy-approved placeholders before the request reaches the AI model — while the sentence stays useful for the AI to act on.
Summarize for the board: patient Maria Lopez (MRN 4471892, DOB 03/14/1968) was started on metformin after her HbA1c result.
Summarize for the board: patient [Person 1] (MRN [ID 1], DOB [Date 1]) was started on metformin after her HbA1c result.
Wants to read /Finance/Q3-board-deck and email it to an external address.
Restricted folder + external recipient
Replace identified PHI or customer data with policy-approved placeholders before an AI request is submitted.
Require human approval when an agent attempts to access a restricted folder, run a high-risk command, or communicate externally.
Record the user or agent, detected risk, policy decision, action taken, and timestamp for review.
One platform, policy packs tuned to your industry
Configurable policy packs adapt protection to regulated and high-value workflows — or apply iDox.ai's detection and policy intelligence to define your own.
Healthcare
PHI, medical codes such as ICD-10, CPT, and HCPCS, patient and claims identifiers, and organization-specific handling policies.
Biotech & life sciences
Research data, trial records, protocols, laboratory results, proprietary formulations, and submission materials.
Legal services
Privileged communications, client-matter identifiers, contracts, filings, case files, and discovery content.
Government & public sector
Citizen data, procurement information, mission-sensitive records, and organization-defined controlled-data categories.
Education
Student records, admissions data, research information, accessibility records, and institutional financial data.
Customer-defined
Beyond the built-in vertical packs, authorized teams can apply iDox.ai's detection and policy intelligence to define fully custom rules — internal identifiers, terminology, prohibited actions, approval requirements, and risk thresholds — so the policy layer evolves with your own regulations and workflows.
Fits your infrastructure and your security stack
Hosting service (cloud)
Rapid deployment with policy distribution and maintenance managed by iDox.ai. Endpoints run on Windows, macOS, and cloud-deployed AI-agent environments.
On-premises
Deploy the management console and data storage within your own infrastructure for physical isolation and maximum compliance control.
APIs, SDKs & webhooks
Event forwarding is designed to connect with your existing security stack:
Specific integrations and platform coverage should be validated during technical scoping.
Define once. Enforce everywhere. Keep improving.
Define
Select data categories, agent actions, users, devices, and risk thresholds.
Deploy
Distribute policies to Windows, macOS, and cloud-deployed agent environments.
Enforce
Allow, sanitize, warn, require approval, or block at the point of interaction.
Improve
Review events, tune rules, and adapt controls as workflows and requirements change.
Recommended deployment path
Technical assessment
Confirm workflows, deployment environments, data flows, integrations, and policy requirements.
Focused pilot
Test safe AI sharing and AI-native threat scenarios in one representative workflow.
Policy configuration
Configure sensitive-data categories, vertical rules, approval requirements, and response actions.
Enterprise rollout
Deploy endpoints, connect operational systems, train users, and monitor outcomes through the management console.
Enterprise product value
Data security
Detect and act on sensitive content at the point of AI use, reducing the likelihood of unauthorized disclosure.
Safe AI adoption
Give employees a governed path to use models and agents without relying only on blanket prohibition.
AI governance
Apply configurable policies, approval steps, audit records, and traceability to AI-mediated data handling and agent actions.
Compliance support
Support organizational compliance programs for privacy and AI-governance requirements; technology controls complement, but do not replace, legal and organizational measures.
Operational scalability
Centrally deploy policies, collect events, and govern AI interactions across employees, devices, and cloud-deployed agents.
Enterprise plans
Hosting service
Initial minimum order: 100 units
Contact us- Gen AI and communication security
- AI agent security
- Full feature management console
- Single Sign-on (SSO) support
On-premises deployment
Initial minimum order: 1 server + 100 units
Contact us- Gen AI and communication security
- AI agent security
- Full feature management console
- Deploy on your cloud
Download Guardrail — free for a limited time
Pick your device and install. No credit card, no setup headache — protection starts the moment it's running.
Prefer the command line?
Power users on Windows can install with a single PowerShell command.
iwr -useb https://idox.ai/install-guardrail.ps1 | iexEnterprise questions, answered
Make enterprise AI safe, private, and governable
Request a demonstration to see the management console and endpoint agent in action on your own workflows.
