Cyber Security Checklist for Organisations Without a Security Team
The controls that stop most real attacks, ranked by protection per unit of effort, plus the AI-specific surface. Written for organisations where one person also does IT.
Markdown. No sign-up, no email.
Organisation: _______________ Date: _______ Completed by: _______________
Ranked by protection per unit of effort, not by framework section number. Sections 1 to 2 stop most of what actually happens. Do those before anything else.
1. The six that do most of the work#
- [ ] Multi-factor authentication on email. Do this first. Email resets every other password
- [ ] MFA on all financial systems
- [ ] MFA everywhere else it is offered
- [ ] Automatic updates enabled where you can tolerate them
- [ ] A written list of every system you run, who owns it, where updates come from
- [ ] Backups: 3 copies, 2 media, 1 offline or immutable
- [ ] A real restore tested in the last quarter, not a "backup job succeeded" message
- [ ] Least privilege, nobody is an administrator by default
- [ ] Leaver access removed same-day
- [ ] Payment detail changes verified out of band. Phone the number you already had, never the one in the email
2. Accounts#
- [ ] No shared accounts, every action attributable to a person
- [ ] Password manager provided and used
- [ ] Default credentials changed on every device and service
- [ ] Access reviewed in the last quarter
- [ ] Dormant accounts disabled
- [ ] Administrative work done from a separate account, not the one used for email
3. Devices#
- [ ] Disk encryption on laptops and phones
- [ ] Screen lock enforced
- [ ] Remote wipe available
- [ ] Endpoint protection installed and reporting
- [ ] Personal devices with company data covered by something
4. Data#
- [ ] You know what personal data you hold and where
- [ ] Not stored where it does not need to be
- [ ] Encrypted in transit, TLS everywhere
- [ ] Retention period defined and actually enforced
- [ ] Test environments do not contain real personal data
5. Suppliers#
- [ ] List of suppliers holding your data
- [ ] What each holds, recorded
- [ ] Their breach notification obligation to you, known
- [ ] Contractor access removed when engagements end
6. AI-specific surface#
- [ ] Approved AI tool provided whose terms exclude training on your data
- [ ] Staff told which tool, and what may not be pasted into it
- [ ] If you built anything with a model: it cannot reach systems beyond its declared tools
- [ ] Destructive or outbound actions require human approval
- [ ] Prompt injection considered. What happens if retrieved content contains instructions?
- [ ] No component holds both broad read and broad write access
- [ ] AI disclosure in place. EU AI Act Art. 50 has applied since 2 Aug 2026
- [ ] Deepfake / voice authorisation defended by out-of-band verification, not by recognising a voice
7. People#
- [ ] Training focuses on process, not spotting bad English. AI phishing is fluent
- [ ] Everyone knows how to report a suspected phishing message
- [ ] Reporting is rewarded, never punished
- [ ] Everyone knows that urgency is itself a warning sign
8. If something goes wrong#
- [ ] One named person with authority to disconnect systems and spend money
- [ ] A named deputy
- [ ] Fallback communication channel agreed, if email is compromised
- [ ] Contact list stored offline. Insurer, legal, IT provider
- [ ] Regulatory notification window known before the clock starts
- [ ] Rule agreed: preserve evidence before rebuilding
Sign-off#
| Name | Date | |
|---|---|---|
| Completed by | ||
| Reviewed by |
Accepted risks:
| # | Not met | Why accepted | Revisit by |
|---|---|---|---|