The essentials¶
Everything in our security policies that actually applies to you, day to day. Five minutes. The long versions live in the document register — this page never contradicts them, it just skips the ceremony.
Your quarterly refresher: open the repo in Claude Code and run
/isms:onboard. It walks you through this page and logs your training automatically. See Get set up.
The twelve rules¶
-
Protect your login. Unique passwords, minimum 12 characters, MFA on everything that supports it — no exceptions for production. (Access control)
-
SSO first. Sign in with SSO wherever it's offered. Separate local logins are only OK for accounts that touch nothing sensitive.
-
Secrets never leave the secret manager. No API keys, passwords or tokens in code, chat, docs, or config files — ever. If you spot one, report it and rotate it. (Cryptography)
-
Your laptop is the office. Disk encryption on, screen lock on, OS updated. It's where customer data lives, so treat it that way. (Remote working · BYOD)
-
See something, say something — immediately. Phishing, a lost device, a weird login, a scammer abusing a feature. Fast beats perfect; reporting is never punished. Slack first (
@-mention #Security, #DevOps, #Development or #General — whichever fits, or #General if unsure), email second (security@soon.works, for the paper trail). (Event reporting) -
AI tools: paid plan, training off, read-only, least data. Claude, ChatGPT/Codex, Cursor and Gemini are approved on paid plans, with connectors to GitHub, Gmail, Drive, Intercom, Stripe, the Soon app and read-only production. Keep training off (on Codex, "Include environments" too), never press the feedback buttons, pull only what the task needs (no bulk exports of customer data), and let a person, not the agent, change production or billing. Never for customers flagged
restrictedExternalProcessors(GXO): their contract forbids it. Never paste secrets. Checked every quarter. (Acceptable use §2.6 · Privacy & data protection) -
Least privilege. Ask for the access you need, hand back what you don't. Production access is personal, MFA-protected, and reviewed quarterly.
-
Everything ships through a PR. Code and policies alike — review is our change control. Nobody pushes to main. (Change management)
-
Install software deliberately. On a machine that touches production or customer data, install from the vendor or an app store — never from a search result or a link someone sent you. Ask Thomas first for anything that (a) needs admin rights or a system extension, (b) can read your screen, keystrokes or clipboard, or (c) signs into a work account. That covers most browser extensions, meeting recorders and clipboard managers. Keep work in its own browser profile: an extension can read every site that profile is signed into. What is already installed is reviewed at the quarterly device check. (Software policy · Device Security Standard)
-
Someone leaving, or a device lost? Access gets revoked the same day — tell Thomas immediately, then sort the rest.
-
Working in public? Lock before you walk. Mind shoulder-surfing on trains, cafés, planes. (Clear desk & screen)
-
A rule doesn't fit reality? Say so. Don't quietly ignore it — raising it is how the system improves, and it's never a problem to flag one.
When something goes wrong¶
- Tell someone now — Slack (
@-mention #Security or #General) and security@soon.works. Don't sit on it. - Olaf coordinates, Thomas handles the technical side.
- If customer data may be involved, we notify affected customers without delay — that's why your speed matters.
Who does what¶
| Olaf | Security lead (ISM), privacy, audits — first stop for questions |
| Thomas | Tech & infrastructure security, access requests |
| Andrea (CyberSquad) | External consultant, runs our internal audit |
Last distilled from the full policy set: 2026-07-02. If this page and a policy ever disagree, the policy wins — and please flag it (rule 12).