AI Cyber Threat Tabletop Exercise Analytics

Executive opening

An AI-enabled cyber incident may not begin with a recognisable malware alert. It may begin with a convincing message, a plausible voice, a compromised supplier, an urgent payment request or several weak signals arriving in different parts of the organisation at once.

That ambiguity changes what leadership needs from an exercise. The question is not simply whether the security team can identify a technical compromise. It is whether the organisation can verify what it knows, move intelligence between departments, authorise action, communicate with confidence and recover when the facts keep changing.

For Australian boards and executives, an AI cyber threat tabletop exercise should therefore be more than a rehearsal of one scenario. It should create evidence about how the organisation behaves under uncertainty—and whether it learns between exercises.

This is where longitudinal tabletop analytics have leadership value. They can help show patterns in authority, trust, intelligence flow, adaptation and recovery over time. OEX (Organisational Exercise Intelligence) and OCI (Organisational Confidence Index) are useful supporting evidence frameworks in that conversation. They are not a magic score, a guarantee of resilience or a substitute for judgement.

The prize is a clearer answer to this question: When an unfamiliar threat moves faster than the plan, how does this organisation make and improve decisions?

AI changes the uncertainty leaders must govern

The Australian Signals Directorate’s Australian Cyber Security Centre (ACSC) assessed that artificial intelligence almost certainly enables malicious actors to operate at greater scale and speed. Its 2024–25 report recorded phishing as an initial-access technique in 38% of reported incidents and described increasingly scalable social engineering.¹

These findings do not mean every incident is AI-generated, nor do they establish a national prevalence rate for deepfakes. They do show why leaders should test the human and organisational controls around technical security.

The exercise is not trying to predict the next attack. It is testing whether the organisation can establish authenticity, escalate concern and act within its risk tolerances when a supplier, executive identity, customer channel or critical third party is uncertain.

Why a single cyber exercise is not enough

A well-run tabletop can reveal a great deal. It can expose unclear authority, weak dependencies, slow escalation, contradictory messages or an untested recovery assumption. But a single session remains a sample of organisational behaviour in a particular scenario, with a particular group of people, at a particular time.

The Australian Institute for Disaster Resilience’s Managing Exercises handbook treats evaluation and lessons management as core parts of exercising. Observations should become corrective actions, not remain a debrief document.²

Longitudinal analytics make the learning question visible. Across comparable exercises, leaders can examine:

whether the same decision bottleneck recurs;
whether escalation and intelligence flow improve;
whether teams challenge assumptions without personal blame;
whether dependencies and recovery become more credible; and
whether previous actions were closed and retested.

These are observations and trends, not causal proof that a score predicts real-incident performance. The value is disciplined evidence for leadership conversations, not false precision.

The leadership capabilities an AI-era tabletop should surface

1. Authority: who can act before certainty arrives?

AI-enabled social engineering can create pressure to act before verification is complete. An organisation that requires several layers of approval for every consequential decision may become slow; an organisation that lets anyone override controls may become unsafe. Resilience depends on knowing which decisions can be made, by whom, under what threshold and with what record.

The Australian Government Crisis Management Framework describes coordination mechanisms that bring relevant stakeholders together and align decisions with expert advice.³ In a tabletop, leadership can observe time to recognise, time to escalate and time to authorise. It can also record decision reversals, evidence cited, dissent heard and unresolved dependencies.

The point is not to reward speed for its own sake. The better question is whether authority supports proportionate action while preserving challenge and review.

2. Trust: can people verify without humiliating the person who raised doubt?

Deepfake uncertainty and impersonation scenarios make verification a cultural issue as well as a control issue. If questioning an executive request is treated as disloyal, staff may comply. If raising a concern triggers blame, the warning may arrive too late.

Psychological safety—the belief that people can raise questions, concerns or mistakes without punishment or humiliation—is associated with speaking up and learning. The Australian Public Service Commission describes it as a foundation for honest communication, ethical decision-making and organisational performance.⁴ It does not mean that every concern is correct or that accountability disappears.

Analytics should therefore avoid scoring individuals as “weak links”. Ask whether the system makes verification normal and whether challenge is treated constructively.

3. Intelligence flow: where does the signal stop?

An AI-enabled incident may produce small, ambiguous signals across security, finance, procurement, legal, privacy, communications, customer operations and an external provider. The risk is not only that one team misses a signal. It is that no team sees the combined pattern.

Exercise observation can map who notices an anomaly, decides it is relevant, combines it with other signals, authorises action and returns the decision to frontline teams and third parties.

The dashboard’s organisational model reflects this chain through identity and trust, structure and roles, decisions and action, output and results, and continual improvement. A technical control can be sound while the surrounding information pathway remains fragmented.

4. Adaptation: can the organisation change its response without losing coherence?

Static playbooks are useful starting points, but an unfamiliar threat can make their assumptions obsolete. The leadership test is whether the organisation can identify which parts of the plan still apply, which need to be adapted and who has the authority to make that call.

Longitudinal exercise analytics can compare maturity, dimension results, trend direction and improvement questions across completed runs. OEX’s six evidence dimensions—robustness, exposure, blind spots, trust, adaptability and recovery—offer a structured vocabulary for that review when sufficient evidence is captured. OCI’s confidence tiers, including “insufficient evidence” and “not captured”, are equally important: an honest gap is more useful than an invented assurance.

It asks whether the evidence base is becoming broader, more consistent and more actionable.

5. Recovery: can critical services continue through dependency failure?

An AI-enabled incident may involve a cloud provider, managed security service, payment platform, identity provider, communications channel or data custodian. A plan that works only while every supplier responds immediately is not a robust plan.

APRA’s Prudential Standard CPS 230 requires relevant APRA-regulated entities to manage operational risk, maintain critical operations within tolerance during severe disruption and oversee material service providers.⁵ CPS 230 does not apply universally to every Australian organisation, but its emphasis on tolerances, continuity and dependencies is a useful leadership reference.

Exercises can test recovery without pretending to certify it. Ask what can continue, for how long, with which degraded controls and under whose authority. Then record whether the answer becomes more credible after retesting.

Australian governance: assurance must be evidence-led

Australian executives should distinguish between a completed exercise and demonstrated preparedness. For APRA-regulated entities, CPS 230 makes operational resilience and service-provider management explicit supervisory concerns. ASIC has also urged directors and financial-services licensees to test operational resilience, crisis responses and third-party vulnerabilities.⁶

Australian Government AI policy materials emphasise accountability, transparency, privacy, security, human oversight and monitoring in their applicable contexts.⁷ OAIC guidance confirms that privacy obligations continue when personal information is handled through artificial intelligence.⁸ Neither source is a universal AI Act or certification that a system is safe. Boards should ask for evidence of learning, not only evidence that an exercise occurred.

How exercise analytics change the conversation

Traditional reporting can lead with attendance, scenario completion and a list of actions. Those facts have value, but they can leave leadership with a binary question: did the exercise pass?

Analytics support a more useful conversation:

From completion to capability: What changed in authority, trust and recovery between runs?
From isolated findings to patterns: Which bottlenecks recur across scenarios or jurisdictions?
From confidence to evidence coverage: Which OEX dimensions are supported, and which are not captured?
From technical ownership to enterprise ownership: Where do finance, procurement, legal, people, communications and suppliers need to act together?
From speed theatre to decision quality: Was action timely, proportionate and open to challenge?
From a report to a learning loop: Were priority actions closed and retested?

Show limits as clearly as the signal: illustrative landing-dashboard histories should not be presented as historical proof. Use evidence-gated measures, completed-run history and clearly labelled observations.

Practical actions for Australian boards and executives

Set questions before scores

Ask what the exercise must reveal about authority, verification, dependencies, recovery and learning. A score without a decision question invites performance management rather than insight.

Exercise verification across functions

Test call-back procedures, dual approval, out-of-band confirmation, payment controls and escalation. Include security, finance, legal, privacy, procurement, communications, operations and material providers; record where signals travel and stall.

Compare like with like, then retest

Note scenario difficulty, participants, exercise mode, scope and evidence coverage. Familiarity can improve performance without proving broader resilience. Close actions with owners and evidence, then revisit the same dependency or decision threshold.

Protect people and evidence

Do not use analytics for punitive individual rankings or to infer mental-health status, trauma or personal vulnerability. Report system patterns and anonymise where appropriate. The proposed lens of threat-informed leadership—inspired by trauma-informed principles of safety, trust, collaboration, empowerment and cultural awareness—is not a validated clinical framework and must never justify surveillance or pathologising employees.

Conclusion

AI raises the cost of unclear authority, fragmented intelligence and cultures that discourage verification. An effective Australian tabletop programme should examine how the organisation resists, responds and recovers; how intelligence moves between departments; how dependencies behave; and whether learning survives beyond the room.

OEX and OCI can structure that evidence, especially when they expose insufficiency rather than conceal it. Their purpose is to make learning—and unfinished work—visible over time, not to replace executive judgement or claim predictive certainty.

The leadership question is: What evidence will your next exercise produce that shows not only how you responded, but how your organisation is becoming better able to respond to an unfamiliar threat?

Key takeaways

AI is increasing the scale and speed of social engineering, but deepfake prevalence should not be exaggerated.
A tabletop is an observation of behaviour in a scenario, not proof of real-incident performance.
Longitudinal analytics can surface recurring patterns in authority, trust, intelligence flow, adaptation and recovery.
Decision velocity should be assessed alongside evidence quality, proportionality, dissent and unresolved dependencies.
OEX and OCI are supporting evidence frameworks; “insufficient evidence” is a legitimate result.
Psychological safety supports verification and learning without removing accountability.
Threat-informed leadership is a proposed management lens inspired by trauma-informed principles, not a clinical or validated framework.
Boards should ask whether exercise actions are closed and retested.

Frequently asked questions

What is an AI cyber threat tabletop exercise in Australia?

It is a facilitated scenario exercise testing how an organisation identifies, verifies, escalates and responds to a cyber threat involving artificial-intelligence-enabled uncertainty. It should include business, governance, human and third-party decisions—not only technical detection.

Can tabletop exercise analytics prove that an organisation is resilient?

No. Analytics can describe observations, trends, evidence coverage and recurring patterns. They do not, on their own, establish causation or guarantee performance in a live incident. Findings should be validated through action closure, retesting and broader operational evidence.

What should a board ask after an AI-enabled cyber exercise?

Ask which decisions were delayed, where intelligence stopped, whether verification was effective, which dependencies were untested, and what changed since the previous exercise.

How should leaders use OEX and OCI?

Use them as supporting frameworks for evidence about robustness, exposure, blind spots, trust, adaptability, recovery and confidence. Treat missing data as a finding, not a failure to hide.

References

1. [Australian Cyber Security Centre, *Annual Cyber Threat Report 2024–25*](https://www.cyber.gov.au/about-us/view-all-content/reports-and-statistics/annual-cyber-threat-report-2024-25) (2025).

2. [Australian Institute for Disaster Resilience, *Australian Disaster Resilience Handbook 3: Managing Exercises*](https://knowledge.aidr.org.au/media/3547/handbook-3-managing-exercises.pdf) (2012, second edition).

3. [Australian Government, *Australian Government Crisis Management Framework: Decision-making and coordination mechanisms*](https://www.pmc.gov.au/resources/australian-government-crisis-management-framework-agcmf/crisis-coordination/decision-making-and-coordination-mechanisms) (current guidance, accessed 2026).

4. [Australian Public Service Commission, *Building psychological safety – whole-of-service approach*](https://www.apsc.gov.au/initiatives-and-programs/workforce-information/research-analysis-and-publications/state-service/state-service-report-2023-24/integrity/building-psychological-safety-whole-service-approach) (2024).

5. [Australian Prudential Regulation Authority, *Prudential Standard CPS 230: Operational Risk Management*](https://www.apra.gov.au/sites/default/files/2023-07/Prudential%20Standard%20CPS%20230%20Operational%20Risk%20Management%20-%20clean.pdf) (2023; effective 2025).

6. [Australian Securities and Investments Commission, *Key issues outlook 2026*](https://www.asic.gov.au/about-asic/news-centre/news-items/key-issues-outlook-2026) (2026).

7. [Australian Government Digital Transformation Agency, *Artificial intelligence policy*](https://www.digital.gov.au/policy/ai) (current guidance, accessed 2026).

8. [Office of the Australian Information Commissioner, *Artificial intelligence and privacy*](https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/artificial-intelligence) (current guidance, accessed 2026).

Resilience in the Age of AI-Enabled Ambiguity