How the Endpoint Sensor Works

A plain-English technical explanation of how AiEGIS captures AI interactions on managed devices — including how it handles encrypted HTTPS traffic.

The Core Question: How Do You See Encrypted Traffic?

This is the right question to ask before deploying any endpoint monitoring tool. Modern AI platforms (ChatGPT, Claude, Copilot, Gemini) all use HTTPS, which encrypts traffic in transit. Here is exactly how AiEGIS captures prompts and responses through that encryption.

Short Answer

AiEGIS does not break or intercept HTTPS at the network level. Instead, it hooks into the application layer — specifically the browser's JavaScript execution environment and native app APIs — where the content is decrypted before it is displayed to the user. The sensor reads at the point where the content is plaintext, not in transit.

Capture Methods

AiEGIS uses three capture methods depending on the deployment environment. Enterprise security teams should evaluate which methods are active in their deployment.

Primary

Browser Extension Hook

A browser extension (Chrome, Firefox, Edge) injects a content script into AI platform pages. The script reads prompt and response content from the page DOM after the browser has decrypted the HTTPS response. No TLS interception. No certificate manipulation.

Enterprise

Local Proxy (Optional)

For enterprise deployments requiring full API-level capture (including native apps), a localhost proxy can be configured with a company-issued CA certificate. Requires explicit deployment by IT with employee notice (Art. 26.7).

Roadmap

Native App API Hooks

For Copilot and enterprise SSO-integrated tools, AiEGIS can hook the native app's API client library. In scope for v0.6.

How Browser Extension Capture Works — Step by Step

Step 1
Employee opens ChatGPT / Claude / Gemini in a browser

The browser loads the page normally over HTTPS. The TLS connection is entirely standard — no interception, no custom certificates.

Step 2
Extension content script activates on the AI platform domain

The AiEGIS browser extension's content script is loaded into the page context. It is scoped to the specific AI platform domains listed in the extension manifest (e.g., chatgpt.com, claude.ai).

Step 3
Employee types a prompt and submits

The content script intercepts the prompt before it is sent — from the input field value, or by observing the fetch/XHR request body through the page's JavaScript context. At this point the content is plaintext.

Step 4
AiEGIS calls /api/protect with the prompt

The extension sends the prompt to the local AiEGIS endpoint (or the cloud API, depending on deployment). All 15 security layers run — PII detection, injection scanning, policy enforcement.

Step 5
Decision: ALLOW, WARN, or BLOCK

If ALLOW: the prompt is sent to the AI platform normally. If WARN: the prompt is sent but logged and surfaced in the IT dashboard. If BLOCK: the prompt is intercepted before transmission — the AI platform never receives it.

Step 6
Response is logged (not blocked)

The AI platform's response is captured from the page DOM after rendering and logged. Responses are not blocked in v0.5 — response enforcement is in scope for v0.6.

What Data Leaves the Device?

Key Privacy Control

Prompt content is sent to AiEGIS for policy evaluation. In self-hosted deployments, this call goes to your own infrastructure — prompt content never leaves your organisation. In cloud deployments, prompts are sent over TLS to AiEGIS servers, processed, and not retained beyond the session log you control.

Data TypeCaptured?Where StoredRetention
Prompt textYes — for policy evaluationYour infrastructure (self-hosted) or AiEGIS (cloud)Configurable, default 90 days
AI response textYes — logged after renderYour infrastructureConfigurable
Employee identityAgent ID (you assign)Your infrastructureConfigurable
TLS keys / certificatesNeverN/AN/A
Passwords / session tokensNever — L10 redacts before loggingN/AN/A
Non-AI web browsingNever — extension scoped to AI domains onlyN/AN/A

Coverage Scope and Limits

What AiEGIS Covers (v0.5)

Managed devices with the browser extension installed on Chrome, Firefox, or Edge. AI tools accessed via browser on those devices.

What AiEGIS Does Not Currently Cover

Personal / unmanaged devices: If an employee uses ChatGPT on a personal phone or home laptop, AiEGIS has no visibility. Organisations should address this through acceptable use policy and device policy — AiEGIS enforces on managed endpoints where it is deployed.

Native desktop apps (Copilot, etc.): The browser extension does not capture native app traffic. The optional local proxy deployment covers this. Native app hooks are on the v0.6 roadmap.

API integrations: If your applications call AI APIs directly (not through a browser), those calls go through /api/protect via the AiEGIS SDK — not the browser extension.

Security of the Sensor Itself

Enterprise security teams routinely ask whether the sensor creates a new attack surface. The honest answer:

Questions?

Enterprise security teams are welcome to request a technical architecture review session, source code review (under NDA), or a penetration test facilitation before deployment. Contact us via the info page.