Stage 3 of 3 · Act
The 90-day plan: what leaders do, ask and decide
Everything in this report comes down to four habits: close the common doors, limit what any one account or agent can do, notice quickly, and recover fast. Here is how to put them in place in 90 days, with an owner for each step.
- 9 min read
- For CEOs, boards and executive teams
- Includes a 10-question self-check
Board questions this part answers
- Which risks have we accepted, and who owns them?
Four layers of defense, built around the business
Pick a layer to see where it stops an attack, what it involves, and who owns it. No single layer is enough; together they turn a breach into an incident.
Tap a layer to see where it stops the attack
- Strong, phishing-resistant sign-in on email, admin accounts and remote access, with no exceptions for executives. 58
- A complete list of internet-facing systems, each with an owner, and critical flaws fixed in days.
- Help-desk resets only after verifying the caller on a channel they do not control. 14
Owned by CISO or Head of IT
- Few administrator accounts, reviewed every quarter, and none shared.
- Separate the person who requests a payment or change from the person who approves it.
- Give AI agents and service accounts only the access their job needs, with hard limits on payments and data leaving the company. 37
Owned by CIO, with the CFO for payments
- Someone watching alerts around the clock, in-house or through a managed security service.
- Pre-approved authority to isolate a device, suspend an account or cut off a supplier at night.
- Records of what each agent and admin did, kept where they cannot alter them.
Owned by CISO, with authority delegated by the COO
- Offline backups of critical systems, tested by actually restoring them into a clean environment.
- A manual way to keep the most important services running for a week.
- A rehearsed incident plan: who leads, who decides, who speaks, and how suppliers keep getting paid.
Owned by COO and CIO
Go deeperWho owns what
An agent can carry out a task. It cannot carry accountability. Neither can a security vendor. These are the owners in most companies; titles differ, the jobs do not.
| Role | Owns | Must be able to show |
|---|---|---|
| Board and CEO | How much risk the company accepts, and which services matter most | Decisions on trade-offs, and funding for the gaps |
| CIO or CTO | Systems, architecture and AI deployments | The agent and system inventories, release checks, restore results |
| CISO or Head of IT Security | Threats, controls and the response team | Coverage figures, detection and containment test results |
| CFO | Payment controls and the cost of disruption | Fraud test results, cash runway with systems down, insurance terms |
| COO and business owners | Each critical service, and what an agent in it may do | Tolerable outage times, manual fallbacks, approved agent scopes |
| General Counsel and DPO | Legal duties and contracts | The notification matrix and supplier obligations |
| HR and Procurement | People and suppliers joining and leaving | Access removed on exit; supplier access reviewed and cut off on demand |
The 90-day plan
Four stages with owners and a proof of done for each task. Tick tasks off as you go. Urgent active threats always take priority over the calendar.
Tap a stage, then tick tasks off as you go
Tasks done: 0 / 16
Know what matters, who can reach it, and where the obvious gaps are.
Rank the 5 to 10 services the business cannot run without, with a tolerable outage time for each.
Owner: CEO, COO and business owners · Done when: A one-page list signed by each owner.
List every AI agent, assistant and connected app, including those bought by departments.
Owner: CIO · Done when: A register with an owner for every entry.
Review who holds admin rights and remove any that are not needed.
Owner: Head of IT · Done when: Before-and-after counts of admin accounts.
Fix or isolate internet-facing systems with critical flaws.
Owner: CISO and COO · Done when: Each one fixed, isolated or formally accepted by a named executive.
Shut the routes the cases in this report used most.
Enforce phishing-resistant sign-in on email, admin and remote access.
Owner: Head of IT · Done when: Coverage percentage and a dated list of exceptions.
Change the help-desk reset procedure and test it with a fake call.
Owner: Head of IT and HR · Done when: Result of the test call.
Add callbacks and a second approver for any new bank details or unusual payment.
Owner: CFO · Done when: An unannounced fraud test, and what happened.
Set hard limits for each AI agent: what it may read, send, change and pay.
Owner: CIO and business owners · Done when: A denied-action test for every agent with real authority.
Review critical suppliers' access and notification terms.
Owner: Procurement and Legal · Done when: Named contacts, cut-off steps and notice periods on file.
Show, not tell, that the business can come back.
Restore one critical service from offline backup into a clean environment.
Owner: CIO and COO · Done when: Time to restore, measured against the tolerable outage.
Run a half-day exercise: ransomware plus a supplier failure, with executives in the room.
Owner: CEO and incident lead · Done when: A list of what failed, with owners and dates.
Stop an AI agent and everything it started, including queued jobs.
Owner: CIO · Done when: A record of the stop test.
Rehearse the 72-hour notification with counsel.
Owner: General Counsel and DPO · Done when: Draft notices ready, and the time it took.
Decide what risk remains, and fund the next round.
Present the self-check results, what was fixed, and what remains open.
Owner: CEO · Done when: A dashboard with exceptions shown, not averaged away.
Accept or fund each remaining gap, with a named owner and a date.
Owner: Board · Done when: Minuted decisions.
Set the routine: monthly exception review, quarterly exercises, re-testing after big changes.
Owner: Executive committee · Done when: Dates in the calendar.
The first 24 hours of an attack
What the leadership team does, in order. Keep a printed copy: the systems holding your plan may be the ones that are down.
Name one incident lead
One person runs the response and reports to the CEO. Everyone else knows who that is.
Contain first, investigate second
Isolate affected systems, revoke stolen access and pause risky integrations and agents. Stopping the visible agent may not stop the work it started; check queued jobs and connected services.
Preserve evidence and start the log
Record what was seen, when, and every decision. Do not wipe and rebuild before evidence is kept; you will need it for insurers, regulators and the police.
Call counsel and the insurer
Their breach lines exist for this. In Indonesia, if personal data is involved, the 3 × 24-hour clock may already be running. 43
Decide how the business keeps running
Manual processes, which channels stay open, and how staff, suppliers and customers are told.
Say only what you know
Confirmed facts, what you are doing, and when you will update next. Do not blame AI, or anyone, before the evidence supports it.
Restart on evidence, not on hope
Bring systems back when they are clean, access is restored to the right people, and transactions reconcile, not just because the screen works again.
The board self-check: ten questions
Answer honestly; 'don't know' is a useful answer. Your result shows where to start and whom to ask. Nothing you enter leaves this page.
Answer each question for your company
Go deeperHow the score works
How the score works: 'Yes, with evidence' counts 2 points, 'Partly' 1, and 'No' or 'Don't know' 0, out of 20. Below 40% is Exposed, 40 to 69% Partly prepared, 70% and above Well prepared. A useful answer has a date and a result: 'We restored this service on this date and met the target' is evidence; 'We have backups' is not.
Go deeperWhat the board should see every quarter
Measures with an owner and a trigger for action. Show the worst items next to the averages; one untested recovery plan can matter more than excellent coverage elsewhere.
| Measure | How it is calculated | What to watch |
|---|---|---|
| Agents with an owner | Production agents with a named owner ÷ agents discovered | Agents found that nobody approved |
| Risky actions under control | High-impact actions with tested approval ÷ all high-impact actions | Include tools and delegated tasks in the count |
| Strong sign-in coverage | Protected privileged accounts ÷ all privileged accounts | Separate people, service accounts and agents |
| Age of critical exposures | Days open for each critical weakness facing the internet | The oldest items, and any accepted exceptions |
| Time to contain | Detection to confirmed removal of harmful access | Include nights, weekends and supplier delays |
| Recovery proven | Critical services restored within target ÷ critical services | Which services have never been tested |
| Supplier readiness | Critical suppliers with a tested escalation route ÷ critical suppliers | A certificate alone is not a test |
Questions leaders ask
What should a board ask about cyber security?
Which outage would hurt us most and how long could we survive it; what each AI agent can do; who owns each supplier connection; whether one person can request and approve a payment; when we last restored from backup; whether finance would stop a fake CEO request; and which risks we have accepted and who owns them.
What should we do in the first 24 hours of a cyber attack?
Name one incident lead, contain affected systems and stolen access, preserve evidence and log every decision, call legal counsel and your insurer, decide how the business keeps running, and communicate only confirmed facts. In Indonesia, start the 3 × 24-hour UU PDP clock if personal data is involved. 43
Who is responsible for cyber security in a company?
The board owns the risk, the CEO owns the response, and a named executive, usually the CISO, CIO or Head of IT, runs security day to day. Business owners decide what their services and AI agents may do.
Where do we start if we have done nothing yet?
Turn on strong multi-factor sign-in for email, admin and remote access; make offline backups and test restoring them; add a callback and second approver for payment changes; and write a one-page incident plan with names and phone numbers.
How do we measure our readiness?
Use the ten-question self-check on this page. 'Yes, with evidence' scores 2, 'Partly' 1, and 'No' or 'Don't know' 0; below 40% is Exposed, 40 to 69% Partly prepared, and 70% or more Well prepared.

