LLM Tracking shows which accounts employees use on web-based AI tools, such as ChatGPT and Claude. It's part of Swif Browser Security, Swif's browser extension.

Employees often use AI tools with personal accounts, such as a personal Gmail address, instead of their work account. Company code and data can then end up in accounts your organization doesn't control. LLM Tracking helps you find this, and act on it.

LLM Tracking works two ways:

Privacy: cookie and token values never reach Swif. To find out which account is signed in, Swif Browser Security reads some of the AI app's own cookies or site data, in the browser. Those values stay in the browser, or go only to the AI app's own server. Swif receives only the account's email address and domain, which app it's on, whether the account is work or personal, the device, and the time.

Swif never collects passwords, prompts, responses, or chat history.

To see AI coding agents and their accounts on the device itself, such as Claude Code or Cursor, use the Swif AI Agent Activity Monitoring Policy.

When an employee opens a supported AI app in their browser, Swif Browser Security detects the account they're signed in to, and records it. It runs in the background, with nothing shown to the employee, and doesn't need a rule.

Automatic account detection works in Swif Browser Security 2.2.38 or later, in Chrome, Edge, and Firefox. It isn't available in Safari yet, on Mac, iPhone, or iPad.

Automatic detections are recorded as Account detected events. They don't appear on the Insight Dashboard, which shows rule-based sessions only.

Each AI app shows its signed-in account differently, so Swif Browser Security reads what each app needs, in the browser:

No cookie or token value is ever sent to Swif.

With a rule, you can track personal accounts on more AI tools, and choose what happens when one is found.

Gemini, Copilot, Perplexity, and DeepSeek are tracked only when a rule includes them.

Swif Browser Security checks the signed-in account, and classifies it:

The notice names the AI tool, and has two options:

The notice doesn't stop the employee from using the tool.

With blocking on, the employee sees a screen that stops them from using the tool with a personal account. To continue, they switch to a work account or sign out.

Automatic account detection never shows anything to employees.

Go to Settings > Team Settings > LLM Account Monitoring. If you haven't created a rule yet, you'll see a prompt to create your first one.

3. Click Create.

You can create several rules, with different tools, work domains, and actions, for different groups.

Click the ⋮ menu on the rule's row, and select Edit or Delete. Deleting a rule stops rule-based tracking for it right away.

Go to Shadow IT > Insight Dashboard, and find the AI Adoption & Usage section.

Both widgets use the time range at the top of the section, Last 30 days by default.

Click a tool to see its sessions by sign-in type, its trend, and the rules that apply to it.

Click a rule to open it in LLM Account Monitoring.

LLM Tracking doesn't need employees to sign in to Swif Browser Security. On enrolled devices, it identifies the device by its Swif enrollment.

Each event includes the app and its domain, the sign-in state and type, the confidence level, the email address and its domain, the device, a session ID, and a timestamp.

No. Swif never reads or sends prompts, responses, or chat history. It records only which account is signed in.

It reads some of the AI app's own cookies or site data in the browser, only to find out which account is signed in. See How the signed-in account is identified. Cookie and token values are never sent to Swif.

Not yet. It works in Chrome, Edge, and Firefox.

No. It runs in the background, and shows nothing.

Automatic account detection records nothing. With a rule, the session is recorded with sign-in type Unknown, and no notice is shown.

Yes, if the employee has allowed Swif Browser Security there. Extensions don't run in private windows by default. Once allowed, automatic detection and rule-based tracking both work as normal.

Each browser has its own setting: Chrome's Incognito, Edge's InPrivate, and Firefox's private windows.

Rule-based tracking stops: no notices, blocks, or rule-based records. Automatic account detection keeps recording signed-in accounts on supported apps.

Yes. Each rule applies to the user groups you choose.

Gemini, Copilot, Perplexity, and DeepSeek are tracked only with a rule. Create a rule that includes them.