A 90-Day AI Cyberattack Plan for CEOs and Boards

The boardroom shift: resilience over perfect prevention

AI-era cybersecurity is a management system, not a product category. It needs to combine prevention, detection, containment, recovery and accountability. NCSC describes cyber resilience for leaders as the strategic foresight to prepare for, respond to and recover from cyberattacks.

That means board conversations should move away from technical inventory and toward business impact:

• Which operations would create material customer, revenue, legal or regulatory harm if they stopped?
• What is the maximum tolerable downtime for each critical operation?
• How long would recovery actually take today?
• Who makes decisions during ransomware, cloud failure, identity compromise or a major supplier outage?
• What would the company say to customers, regulators, employees and investors?

AI adds a second layer. The company must manage not only attackers’ use of AI, but also its own adoption of AI. WEF’s guidance for senior risk owners calls for assessing the vulnerabilities AI adoption may create, evaluating business impact, identifying controls and balancing residual risk against expected benefits.

Seven decisions to make in the next 90 days

1. Put cyber risk permanently on the board agenda

Cyber risk should not appear only as a quarterly technical update. It belongs in the same risk conversation as liquidity, safety, regulatory exposure and operational continuity. NCSC explicitly links cyber incidents to operational, reputational, financial and legal consequences. Analysis shared by the Harvard Law School Forum on Corporate Governance also notes that corporate disclosures on cyber and AI oversight have become more common and more robust.

The board does not need a count of firewalls. It needs to see the business services at risk, unresolved critical vulnerabilities, privileged access exposure, detection and recovery performance, supplier dependencies and AI-use controls. WEF also frames AI risk tolerance as a top-leader question: has the organization set and communicated the risk parameters for adopting AI technologies?

2. Put numbers on critical operations and recovery

AI may accelerate attacks, but the board’s core responsibility remains the same: protect the organization’s most important work. For each critical operation, leadership should know the maximum acceptable outage, the fallback process, the recovery owner and the communication path.

Backups are not enough because a backup that cannot be restored under pressure is only a comfort blanket. Directors should ask whether backups are isolated, whether recovery has been tested, and how many hours it took to restore real systems. NCSC’s resilience message is clear: organizations must prepare for events they cannot fully control.

3. Use AI in defense, but judge it by outcomes

If attackers use generative AI to move faster, defenders cannot rely only on manual processes. AI-assisted monitoring, alert triage, investigation support and prioritization may all have a role. But the board should not measure success by whether an AI security tool has been purchased.

The test is operational: did time to detect improve? Did containment get faster? Are false positives manageable? Are audit logs complete? Who approves exceptions? WEF’s AI cyber-risk approach emphasizes identifying controls and balancing residual risk against business benefit, which is the right lens for defensive AI as well.

4. Govern the company’s own AI and data

The company’s internal AI use can create risk even when no attacker is present. Sensitive data may be copied into tools that are not approved. Applications connected to large language models may have excessive permissions. AI outputs may be hard to audit. Training or retrieval data may be changed without enough control.

WEF calls for organizations to assess the vulnerabilities and business impacts created by AI adoption and to identify the controls needed to manage residual risk. A joint cybersecurity information sheet published through a US defense channel says securing data used in AI and machine-learning systems is important to the accuracy and integrity of AI outcomes, with a focus on sensitive, proprietary or mission-critical data.

Boards should require clear rules for internal AI systems, external AI tools, LLM-connected applications and the circumstances under which confidential information may be used. Access controls should cover the data that feeds AI systems, including training data, prompts, logs, retrieval datasets, embeddings and other model-related assets.

5. Connect CISO authority to executive authority

A chief information security officer cannot be accountable for AI-speed risk while lacking the authority to stop dangerous exceptions or secure the budget to fix known exposure. If cyber risk is a boardroom priority, as NCSC says, the CISO needs a route to explain business risk directly to senior leaders and directors.

The board should confirm whether the CISO can challenge risky system changes, review AI deployments, require business units to remediate critical vulnerabilities and influence third-party risk decisions. WEF’s AI cyber-risk approach also puts risk, controls and residual risk in a senior management frame rather than treating them as narrow technical issues.

6. Treat cloud and suppliers as resilience dependencies

AI cyber readiness does not stop at the company’s network boundary. NCSC has highlighted the need to raise resilience across supply chains. SaaS providers, managed service providers, cloud platforms, development partners, data processors and major subsidiaries should all be mapped as critical dependencies.

For key vendors, contracts should address authentication requirements, access to logs, breach-notification timelines, AI-use policies, data protection, backup responsibilities, audit rights and cooperation during recovery. If a vendor’s notice or logs arrive too late, the company’s own containment and recovery decisions may be delayed.

7. Replace incident-count metrics with resilience metrics

Counting incidents alone does not tell directors whether the organization is getting stronger. A board dashboard should show whether the company can find, limit and recover from harm fast enough.

The CEO’s first 90 days

Days 0–30: Make the risk visible

1. Schedule a board-level review of AI cyber risk.
2. Identify critical operations and confirm the maximum tolerable downtime and actual recovery capability for each one.
3. Ask the CISO to assess both AI-enabled attacks and the company’s own AI adoption risk.
4. Stop confidential data from being entered into unapproved AI tools and create an exception-approval process.

Days 31–60: Strengthen controls and recovery capability

5. Prioritize stronger authentication and access controls for executives, administrators and privileged accounts.
6. Run a ransomware-oriented recovery test from isolated backups. WEF’s 2025 outlook identifies ransomware as a continuing top concern.
7. Review key vendors and contract terms for logs, breach notice, AI use, data protection and recovery support.

Days 61–90: Test decisions and make metrics routine

8. Run an executive tabletop exercise involving the CEO, CFO, legal, communications, the CISO and business-unit leaders.
9. Add detection time, containment time, recovery time, unresolved critical vulnerabilities and AI-use reviews to the board dashboard.
10. Revisit the CISO’s reporting line, stop authority, AI-review authority and budget, then make follow-up a standing board item.

Questions directors should ask every meeting

A board does not need to run the security operations center. It does need to ask questions that force preparedness:

• Have we set the right risk tolerance for AI technologies, and is it understood by decision-makers?
• If a critical operation stops, who starts customer, legal and communications activity, and within how many hours?
• For AI and machine-learning systems, who can change the data, and how do we verify its integrity?
• Can the CISO escalate dangerous exceptions and unresolved risk directly to executive leadership and the board?
• If a major supplier is breached, how quickly would we know, and what fallback could we use?

The goal is not to promise zero breaches. It is to make sure the company can detect trouble early, limit damage, keep essential operations moving and communicate with stakeholders from a position of facts. The CEO and board should not try to choose every security tool. Their job is to set risk tolerance, give the CISO authority and resources, govern AI use, and repeatedly test whether the organization can actually recover.