Data Handling Details

The full detail behind the Security page.

For security reviewers: what Team sends to TorchTiger and when, what is never sent, what stays on your device, and why each permission is needed. In TorchTiger Engineer, the Community edition sends TorchTiger nothing: no TorchTiger backend, no telemetry.

Request a Security Review

How Data Flows

Last updated 2 October 2026

Your Browser

TorchTiger Engineer runs here, only on your ServiceNow instance pages. It keeps your key encrypted and replaces detected personal data in chats before they are sent.

Both editions
Your ServiceNow Instance

It reads with your own session and permissions. Nothing is written until you approve.
Production is locked out: no chat, no AI reads, no writes. The same applies to any instance it cannot confirm is non-production; the extension only reads the instance’s settings to tell production apart.
Both editions
Your AI Provider

Community: Anthropic, under your own key. Team: once a member’s seat is verified, the Anthropic API, Amazon Bedrock or your gateway, per your team’s AI profile. With an Anthropic key, Claude may also run web searches, which Anthropic carries out.
Team only
TorchTiger’s Servers

api.torchtiger.com receives sign-ins, seat checks, check-ins and change records. Your team sees the results in two web apps at app.torchtiger.com. The Engineer web app shows the change log and change dashboard; the Hub manages members, seats and billing.

Where data goes from the extension. The At a Glance table on the Security page is the text version of this diagram.

What Team Sends to TorchTiger

Community sends TorchTiger nothing. On Team, the extension sends these to TorchTiger’s servers at api.torchtiger.com, which also hold your team’s members, roles, policies and the instances you register.

Everything the Team edition sends to TorchTiger, and when.
Sent	What it carries
Sign-inWhen a member signs in	Your work email and password, and an emailed one-time code (on by default). TorchTiger’s servers record each sign-in’s IP address and browser.
Seat checkWhen you sign in, unlock, open the panel or return to a ServiceNow tab	An install ID with your session. Opening the toolbar popup also fetches the team policy.
Check-inEvery 15 minutes	The same install ID.
Change recordAbout 3 seconds after each AI write	The instance address and environment, table, record ID, action, update-set name and ID, and time. The update-set name is a short title the AI writes for the plan. It also carries the record’s name as written, such as a business rule’s name, up to 200 characters and not redacted. If the AI changes a data record such as an incident, its short description or number is sent instead. It also names the record’s type and the chat turn that made the change. If a change is undone, a short revert record follows. TorchTiger’s servers add who made it.

Switching it off: There is no setting to turn change records off.
Who sees it: In the Engineer web app, owners, admins, billing admins and read-only auditors see every member’s change records, including the member’s email.
How long it’s kept: Change records are append-only and are not deleted automatically today.
Offline: A browser that cannot reach TorchTiger for 72 hours falls back to Community rules until it can.

What Stays on Your Device

Your Key, Sealed

Your key is stored encrypted under your passphrase (PBKDF2 and AES-GCM), and the passphrase is never saved. Unlocked, the key sits in memory-only session storage and never reaches the page.

Personal Data, Replaced First

Personal data it detects in your messages and ServiceNow results is swapped for placeholders in your browser before sending. The map back stays on your device.

A Local Change Log

Every AI change is logged in your browser, with the undo copies needed to revert it. It stays on this device, with no sync.

The log is not encrypted, and it survives Reset.

What Redaction Catches

Detected: Emails, card numbers, IP addresses, North American phone numbers and many names.
Not detected: UK National Insurance numbers, postcodes and UK-format phone numbers.
Not redacted: Attached images and PDFs, and Team change records.

Browser Permissions

Asked For at Install

storage
activeTab
scripting
identity
alarms
storage.managed_schema
api.anthropic.com
*.service-now.com instance pages

Never Asked For

No cookies or broad tabs. It cannot read your logins or see your other tabs.
No <all-urls> at install. It holds only api.anthropic.com and ServiceNow instance pages. TorchTiger’s servers and your AI provider are asked for on your click, one address at a time. The manifest lists them as optional, including https://*/*. The grant for TorchTiger’s servers stays after sign-out.
No remotely hosted code. Every line ships in the reviewed package, under a strict Manifest V3 content security policy.
No chrome.storage.sync. Nothing syncs through your browser account.
No third-party analytics or fingerprinting.

Why Each One Is Needed

Each permission and host the extension asks for at install, and why.
Permission	Why
`storage`	Your encrypted key and chats, settings, the instances you’ve opened and the local change log, on this device only, with no sync. On Team, also your install ID and seat token.
`activeTab`	Label the ServiceNow instance you’re on, and scope write mode to it.
`scripting`	Inject the “What’s redacted” editor, only on your click, on the active tab.
`identity`	On Team, get a short-lived token from your identity provider for a keyless Amazon Bedrock connection. It also opens the sign-in page for an organisation set up with single sign-on.
`alarms`	Re-lock your key when the unlock window ends; on Team, schedule the 15-minute check-in.
`storage.managed_schema`	Let your IT pre-set your organisation’s name and Hub link by browser policy. Read-only, and read on every install.
`api.anthropic.com`	Direct calls to your Anthropic account, under your key (Community).
Content scripts	On `*.service-now.com` instance pages only. ServiceNow’s sign-in, portal and support sites are excluded.

Check It Yourself

Readable Code

Our own code ships unminified, with no build step, so your team can read the shipped files. Two third-party PII (personal data) libraries are minified.

Checked on Every Build

Our CI fails the build if code adds third-party analytics or fingerprinting, or hard-codes a host outside a fixed list.

Readable, Not Open Source

Being readable does not make it open source: all rights are reserved. The full source is available to security teams under NDA.

Procurement & Compliance

DPA on Request

A DPA is available on request. Community is serverless and accountless, so TorchTiger is not a data processor for it. On Team, TorchTiger processes your members’ account and billing details, sign-ins, install IDs and change records, including record names as written.

SOC 2 — Not Yet Certified

TorchTiger is not SOC 2 certified yet. We say so plainly and never present it as a current attestation.

Source Access Under NDA

Security teams can read the source under NDA: a named owner (the TorchTiger Engineer maintainer), acknowledged within 3 business days.

Responsible Disclosure

Report vulnerabilities privately to security@torchtiger.com — acknowledged within 3 business days, coordinated disclosure, and good-faith safe harbour.

Bring Your Security Team

Got a questionnaire, review, or data-flow question? We'll walk your security team through exactly how it works.