The proposed exercises raise a practical question for financial institutions: how to restore trusted services when prevention has failed and the scope of a compromise is still uncertain.
G7 finance officials have agreed to develop exercises addressing AI-related cyber threats, Reuters reported on October 16, citing Japanese Finance Minister Satsuki Katayama. The proposed tabletop simulations could involve private financial institutions, depending on the scenario. An agreement to organise drills is not evidence that those drills have already been completed.
The value of such an exercise depends on the decisions it forces participants to confront. A presentation about a sophisticated adversary may raise awareness without testing whether an institution can authorise isolation, establish reliable communications or determine when it is safe to reconnect. A useful simulation exposes uncertainty in those processes before an actual crisis makes experimentation more expensive.
There is already a cross-border testing framework
A US Treasury statement issued on July 31 says the G7 Cyber Expert Group concluded a separate cross-border coordination exercise on May 18, 2026. It simulated a large cyberattack affecting all G7 jurisdictions and brought together finance ministries, central banks, bank supervisors and market authorities. The work examined response, recovery and crisis communications.
That earlier exercise should not be confused with the newly reported AI initiative. It does, however, establish that the new discussion sits within an existing programme of coordination. Treasury also described a longer-term strategy to make exercises more frequent and consistent. The relevant next question is which assumptions the AI-focused scenarios will change and how improvements will be tested afterwards.
An incident response plan must reach beyond the security team
NIST’s April 2025 revision of SP 800-61 integrates incident response into cybersecurity risk management using the Cybersecurity Framework 2.0. Its stated purpose includes better preparation, reduced incident impact and more effective detection, response and recovery. That is a broader objective than discovering an intrusion quickly and passing the alert to another team.
Consider a hypothetical bank whose payment service becomes unreliable while administrators also receive suspicious requests to change privileged access. The exercise is not useful merely because participants recognise both events as alarming. It needs to establish who can suspend affected functions, which business processes must continue, how customers will be informed and what evidence is needed before normal service resumes. These are proposed test questions, not a description of an actual G7 scenario.
The same hypothetical can test conflicting incentives. Operations staff may want an immediate restart; investigators may want to preserve evidence; executives may be concerned about disruption spreading to counterparties. A realistic exercise should make those tensions visible and identify a decision process, rather than assuming that every team will spontaneously choose the same priority.
Recovery is a question of trust as well as availability
Bank of England Deputy Governor Sarah Breeden emphasised the need for recovery capabilities, including rebuilding from a clean foundation, in remarks reported by Reuters on October 15. Her intervention challenged the assumption that defensive systems alone can make a financial institution sufficiently resilient.
In the hypothetical exercise, restoring a server would not answer whether the records it contains are trustworthy. Nor would a functioning login screen establish that the credentials behind it are safe to reuse. A recovery test could therefore ask participants to explain how they would validate data, determine the scope of compromised access and document the basis for reconnection. The point is to require evidence for each decision, not simply a declaration that systems are back online.
Another useful distinction is between restoring technical components and restoring a service. A component may work while an essential dependency remains unavailable. An exercise designed around a meaningful business outcome would reveal that difference; one designed only around a list of servers could miss it. The criterion should be a service that can be used safely, not the number of machines that have restarted.
The AI element needs a defined role
For analytical purposes, an AI-focused scenario should specify what the technology changes. Does it alter the volume of misleading messages, the speed of reconnaissance, the credibility of an impersonation or a defender’s dependence on automated recommendations? Those possibilities require different tests. Merely adding the term AI to an established scenario would not demonstrate that a new risk has been examined.
A defensive system’s output should also be open to challenge within the exercise. Participants could be asked how they would verify a high-confidence recommendation when other evidence conflicts with it. This would test judgement and escalation rather than rewarding uncritical acceptance of an automated answer. It would also avoid assuming that the newest available technology is necessarily the most reliable source during an incident.
Measure what changes after the drill
The most valuable output would be a set of specific findings with owners, deadlines and a subsequent retest. For example, a simulated delay in reaching a counterpart could lead to a verified alternative communications procedure. An unclear recovery decision could lead to an explicit authority and evidence requirement. These are practical improvements against which the next exercise can be compared.
NIST’s risk-management approach provides a useful reference point: preparation and recovery belong within continuing organisational management, not only within the moments when an alert is active. The proposed G7 work will be most meaningful if it follows that logic and tests whether the lessons survive beyond the exercise room.
The announcement is therefore a starting point, not a certificate of resilience. The test ahead is whether financial institutions and authorities can demonstrate that they know how to make difficult decisions together, restore trustworthy services and correct the weaknesses that the simulations uncover.
