A tabletop exercise cannot patch a server, rotate a stolen credential or restore a payments database. It can reveal something that becomes equally important during a systemic incident: whether ministries, central banks, supervisors, market authorities and private institutions know who must decide what, which facts can be shared and how recovery priorities change when several countries are affected at once.
G7 finance ministers and central-bank governors have agreed to develop exercises around AI-enabled cyberattacks, according to Reuters. The simulations are intended to test response protocols and inform revisions to security manuals, with private financial institutions potentially taking part. The initiative extends a programme of cross-border exercises rather than starting from zero.[1]
The exercise tests coordination, not just technology
Questions a cross-border tabletop can expose
| Decision point | What participants must establish | Why AI changes the pressure |
|---|---|---|
| Incident classification | Whether several failures are connected and how severe the event is | Automated attacks can move faster and produce more simultaneous signalsSource 3 |
| Information sharing | Which indicators, exposures and service dependencies can be shared, with whom and when | Machine-speed exploitation reduces the value of delayed reportingSource 2Source 3 |
| Authority | Who can order containment, market measures, public communication or emergency access | Responsibilities are split across firms, supervisors and jurisdictionsSource 2 |
| Recovery order | Which services must return first and how integrity is verified | Common providers can create correlated disruption across institutionsSource 3 |
| Public communication | How to explain uncertainty without creating panic or false reassurance | Synthetic content and rapid rumours can complicate the information environmentSource 1Source 3 |
The G7 Cyber Expert Group’s May 2026 exercise simulated a large-scale attack across every G7 jurisdiction. It brought together finance ministries, central banks, bank supervisors and market authorities, building on earlier work in incident response, recovery and crisis communication. The group then adopted a longer-term strategy to make simulations more frequent and consistent.[2]
AI can magnify ordinary weaknesses
The IMF’s analysis argues that the main systemic danger is not necessarily an entirely new class of attack. AI can accelerate vulnerability discovery, exploitation and adaptation across technologies that many institutions share. A weakness in one widely used provider can therefore produce correlated failures more quickly than human coordination can absorb.[3]
What the current programme can and cannot establish
| Measure | Value | Context |
|---|---|---|
| Can test | Communication paths, escalation rules, hand-offs and recovery priorities | Participants can discover that a contact list, legal authority or dependency map is unusable before a real emergency.Source 2 |
| Can test | Whether protocols still work under faster, wider disruption | The new G7 work is explicitly framed around AI-enabled attacks.Source 1 |
| Cannot prove | That a bank’s technical controls will resist a live attacker | A discussion exercise is not a penetration test or red-team engagement. |
| Cannot prove | That recovery systems contain clean, restorable data | Those capabilities require technical drills and evidence from real systems. |
| Cannot guarantee | That participants will behave the same way under market stress | Real incidents bring incomplete information, legal risk and political pressure. |
A good scenario should force trade-offs
The weakest tabletop confirms that everyone can read a prepared plan. A stronger one introduces conflicting evidence, unavailable personnel, a compromised cloud provider, cross-border legal constraints and a choice between restoring service quickly and preserving forensic certainty. It should force participants to state which losses they will tolerate and what evidence they need before reconnecting a system.
AI also creates a useful design constraint for the exercise itself. The scenario should not assume that every alarming output is authentic, or that an AI-generated alert is correct. Defenders may face generated phishing, automated exploitation, synthetic public claims and their own defensive models producing false positives at the same time.
Financial stability depends on more than each bank being individually secure. Payments, clearing, cloud services, identity systems and market confidence connect institutions across borders. A tabletop cannot build those defences. It can show whether the organisations responsible for them will recognise the same emergency soon enough to act as one system rather than a collection of isolated manuals.
Sources & further reading
- G7 to launch exercises to guard against AI cyber attacks Reuters · 16 Oct 2026 · Accessed 16 Oct 2026
- G7 Cyber Expert Group 2026 Cross Border Coordination Exercise HM Treasury · 31 Jul 2026 · Accessed 16 Oct 2026
- Artificial Intelligence and Cybersecurity in the Financial Sector International Monetary Fund · 30 Jun 2026 · Accessed 16 Oct 2026