Question 1

What is FlowGuard AI?

Accepted Answer

FlowGuard AI is an endpoint-based enterprise AI control platform. We build software that installs directly on employee devices—as a lightweight endpoint agent or browser extension—to monitor, detect, and block sensitive data before it is sent to generative AI services like ChatGPT, Claude, Gemini, or GitHub Copilot. Unlike network proxy or cloud-gateway approaches, our endpoint agent operates at the device level, which means it works regardless of the network the employee is on: office, home, mobile hotspot, or VPN. Our mission is to make generative AI safe to use inside the enterprise.

Question 2

Why endpoint-based? What's wrong with network proxy or SSL inspection approaches?

Accepted Answer

Network proxy and SSL inspection solutions sit between the corporate network and the internet, but they have three fundamental blind spots for AI security: (1) Remote and hybrid workers on home Wi-Fi or mobile data bypass the corporate proxy entirely—the AI traffic never passes through it. (2) IDE-integrated AI coding assistants like GitHub Copilot communicate through the IDE process, not the browser, so proxy-based tools cannot inspect those requests. (3) SSL inspection requires a man-in-the-middle certificate on every device, which raises serious privacy, legal, and performance concerns. FlowGuard AI's endpoint agent solves all three problems by operating directly on the device, intercepting AI-bound data at the application layer before it ever reaches the network.

Question 3

How does the FlowGuard AI endpoint agent work technically?

Accepted Answer

The FlowGuard AI endpoint agent is a lightweight process that runs on Windows and macOS. It hooks into the browser and IDE at the application layer to inspect outbound AI prompt content before transmission. When an employee submits a prompt to any monitored AI service, the agent analyzes the text using a local policy engine—no raw prompt content is sent to FlowGuard servers. If sensitive data is detected (PII, source code, credentials, financial data, etc.), the agent enforces the configured policy: block and alert, mask the sensitive portion, or log the event silently for compliance. The browser extension variant achieves the same coverage for browser-based AI tools without requiring a full endpoint agent installation.

Question 4

What generative AI services does FlowGuard AI cover?

Accepted Answer

FlowGuard AI covers 40+ generative AI services including ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Microsoft Copilot, GitHub Copilot, Cursor, Windsurf, Perplexity, Notion AI, and many others—across both web browser interfaces and IDE plugins. Our shadow AI detection database catalogs 13,000+ AI tools, so even newly emerging services are flagged automatically when an employee accesses them for the first time. Coverage is updated continuously as new generative AI services launch.

Question 5

How is FlowGuard AI different from traditional DLP solutions?

Accepted Answer

Traditional DLP was designed for email, USB drives, and file transfers—it has no understanding of AI prompt context. FlowGuard AI is purpose-built for generative AI security: (1) It understands conversational AI prompt structure, not just keyword patterns. (2) It is endpoint-based, so it covers remote workers and IDE-based coding assistants that proxy DLP cannot reach. (3) It provides AI-specific visibility: which AI tools are used, by whom, how often, and what categories of data were attempted. (4) It requires no SSL inspection certificate or network infrastructure changes—deployment takes minutes, not months.

Question 6

What types of sensitive data does FlowGuard AI detect?

Accepted Answer

FlowGuard AI detects: personally identifiable information (PII) including names, emails, phone numbers, and Korean resident registration numbers (주민등록번호); financial data such as account numbers and credit card details; source code, API keys, and database credentials; internal documents, contracts, M&A materials, and strategy files; healthcare and medical records; and custom data patterns unique to your organization. Detection is optimized for both Korean and English content, with AI-context awareness that reduces false positives from generic text that happens to resemble sensitive data.

Question 7

Does the endpoint agent affect employee privacy or performance?

Accepted Answer

Privacy: FlowGuard AI inspects only the content of outbound AI prompts—it does not record general browsing, keystrokes, emails, or any other activity. Raw prompt text is analyzed locally on the device by the policy engine and is not sent to FlowGuard servers. Only policy violation metadata (timestamp, user ID, data category, action taken) is logged for the security team. Performance: The endpoint agent is designed to be lightweight. CPU usage during prompt analysis is typically under 1%, and the inspection adds less than 100ms latency to AI interactions—imperceptible to end users.

Question 8

How is FlowGuard AI deployed across an enterprise?

Accepted Answer

FlowGuard AI supports two deployment modes: (1) Browser Extension — installable from the Chrome Web Store or deployed silently via enterprise browser policy (Google Workspace, Microsoft Intune). Covers all browser-based AI services. Deployment takes under 5 minutes per device, or minutes across the entire fleet via policy push. (2) Endpoint Agent — an installable application for Windows and macOS that covers both browser-based AI and IDE-integrated coding assistants. Can be distributed via MDM platforms (Microsoft Intune, Jamf, Kandji). No network infrastructure changes, no firewall rules, no SSL interception certificates required.

Question 9

What is Shadow AI and why is it a security risk?

Accepted Answer

Shadow AI refers to generative AI tools that employees use without IT or security team approval—similar to how 'shadow IT' described unsanctioned cloud apps. The risk is significant: an employee might use a free, unvetted AI chatbot that stores conversation history indefinitely, shares inputs with third parties for model training, or lacks enterprise-grade security controls. FlowGuard AI's endpoint agent detects AI service access at the device level, matching it against a database of 13,000+ known AI tools. Every newly detected AI service is automatically logged and flagged, giving security teams a complete map of AI usage across the organization—including tools the IT team never knew existed.

Question 10

Is FlowGuard AI compliant with data privacy regulations like GDPR and Korea's PIPA?

Accepted Answer

Yes. FlowGuard AI is designed to support compliance with GDPR (EU), PIPA (Korea's Personal Information Protection Act), and other regional data protection frameworks. Because the endpoint agent analyzes prompt content locally and does not transmit raw prompt text to FlowGuard servers, it avoids the secondary data processing concerns that cloud-inspection solutions face. The platform provides tamper-proof audit logs, role-based access controls, and data residency options that compliance and legal teams require for AI usage governance.

Question 11

What is MCP (Model Context Protocol), and why does it matter for security?

Accepted Answer

MCP is an open protocol that standardises how AI models connect to external data and tools, so an assistant can reach internal wikis, databases, ticketing systems, and code repositories through one consistent interface. It matters for security because AI stops being something that only exchanges text: it reads from and acts on real internal systems. Data can move even when nobody types anything sensitive, so inspecting prompts alone no longer covers the risk.

Question 12

What are AI guardrails, and how do they differ from AI DLP?

Accepted Answer

AI guardrails are the control layer that inspects and constrains an AI system's inputs, outputs, and actions against policy. AI DLP sits in the first of those layers: it inspects what goes into the model and blocks sensitive data before it leaves the device. Guardrails extend the same idea to what the model returns and, critically, to what an agent is about to do — which tool it will call, against which system, with what arguments. AI DLP is a guardrail; it is not all of them.

Question 13

Can prompt-level AI DLP protect against AI agents and MCP integrations?

Accepted Answer

Not on its own. Prompt DLP inspects what a person typed. In an MCP environment an agent can pull internal data and pass it outward through tool calls without anyone typing a single sensitive character — a request as ordinary as "summarise last week's churn" can end with customer records summarised into an external channel. Prompt DLP remains necessary for the chat surface; it is simply no longer sufficient once agents can act.

Question 14

Does FlowGuard AI provide MCP visibility and agent action guardrails today?

Accepted Answer

Not yet. What ships today is the prompt layer: visibility into which AI tools are actually in use, sensitive data detection and blocking in prompts and responses, Shadow AI discovery including code assistants, policy-based control, and audit logs. MCP connection visibility and agent action guardrails are stated direction, still in development — we describe them separately from the product for that reason. Contact us for timelines and scope.

Generative AI Security
FAQ

Still have questions?

Generative AI SecurityFAQ