G7 AI Tabletop Exercises: Where Financial Information Sharing Can Stall

1. An agreement to rehearse across borders

On October 16, Finance Minister Katayama said future AI-risk tabletop exercises would follow the previous day’s G7 and other meetings.[1] Tabletop exercises rehearse responses through scenario discussion.[2] The agreement offers an opportunity to test decisions together. Plans should be assessed separately from exercise dates, detailed scenarios, participating firms and subsequent results.

For finance, an important difficulty is that the authority learning about a disruption, the authority requesting a report and the party legally able to receive information need not be the same. A bank may be investigating its services while several authorities ask different questions. If updates stall, market participants may have to make funding and trading decisions with important gaps unresolved. SG Group sees the agreement as an opportunity to test these institutional boundaries, rather than as a contest between AI systems. A common document alone is a poor measure of success. The more useful question is which decision lacked which information, and what prevented it from reaching the right recipient.

2. AI adds pressure to an established coordination problem

International exercises and better reporting are established ideas. The G7’s 2020 elements are non-binding exercise guidance. The FSB’s April 13, 2023 recommendations address reporting burden and criteria, timely reporting culture, definitions, secure communication and cross-border sharing constraints. These are background materials, not laws enacted at the October 2026 meetings.[2][3]

AI should not be used to rebrand every existing task as a new problem. The practical question is whether established communication and decision procedures still work when the available time changes. A September 9, 2026 paper from the BIS Financial Stability Institute discusses defensive opportunities alongside shorter remediation windows and third-party dependencies affecting financial institutions. [4] Technical capabilities can change faster than national legal arrangements. As that gap widens, an exercise needs to do more than add a new attack label: it should reveal which reporting steps can be accelerated and which constraints remain.

3. SG Group View: align reporting without erasing its purposes

Multiple reports about one incident can contain both wasteful duplication and necessary differences. An official assessing disrupted services needs a different account from one assessing financial-market spillovers. The first may seek technical records, while the second needs to know which payments or funding transfers are affected. Standardising every field before examining its purpose could produce a shorter form that omits information needed for a decision. Fields serving the same purpose but using different names or ordering, by contrast, offer a clearer opportunity for alignment.

A useful distinction is between facts that can be reused and additions required for a recipient’s particular purpose. A shared core might identify the service, observation time, scope of impact and expected next update. That is an analytical illustration, not a new format adopted by the G7. Recipient-specific sections would retain the explanations required by each framework. The benefit of alignment is better assessed by whether it cuts repeated transcription without removing decision-relevant information than by simply counting deleted fields. Preserving different regulatory purposes can be compatible with reducing the reporting burden.

4. Different reporting triggers change the meaning of silence

Consider a hypothetical disruption affecting services in two jurisdictions. One reporting framework might focus on service impact and seek an early notification; another might require further assessment of materiality. A report in the first jurisdiction and silence in the second would not establish that the second was unaffected. The presence of a report depends on the reporting trigger as well as the incident. This is an illustration of how different conditions shape visibility, not a comparison of actual national thresholds or deadlines. The conditions producing information need examination before its volume is compared.

An international exercise could test how participants interpret that silence. Can a recipient distinguish a threshold not yet met from an assessment still under way or a communication failure? Senders also need to understand, under their own frameworks, what they may share before a formal reporting obligation begins. Urging people to communicate faster without resolving that question leaves them between inappropriate disclosure and excessive caution. Success should not depend on someone improvising helpfully. Participants should be able to explain what information can be transmitted on what basis, reducing the temptation to read silence as evidence of safety.

5. Make an initial report useful and revisable

Waiting for a known cause can cost recipients valuable time; presenting an early hypothesis as settled fact can make correction costly. Initial and subsequent reports therefore need different jobs. The first should communicate changes relevant to decisions at that point, while later updates deepen the investigation and impact assessment. Separating these stages does not make national deadlines or recipients identical. The challenge is to connect framework-specific obligations with the updating of a shared factual account.

Imagine a delay in a payment-related service. An initial report may identify the service and observation time while leaving unresolved whether delayed instructions have been lost or can be processed later. A subsequent answer changes the options available to users. The update needs to show what it replaces in the earlier account, rather than merely add another figure. If old and new versions circulate together, participants in the same meeting can reason from different situations. Assessing the speed of initial reporting should therefore include the reliability of the correction process.

6. The same term can describe different units

International information sharing requires a distinction between the number of reports and the reach of a disruption. Suppose three banks separately report a failure at one external service. There are three reports, but there may be only one underlying failure. Conversely, one bank’s report could cover separate problems with payments, customer notices and internal reconciliation. Adding up reports alone can count a common cause several times or compress several effects into one. Before comparing totals, an exercise should establish what counts as one incident and which relationships need their own record.

More detailed categories are not a complete solution. An initial report may arrive before the cause is known or before authorities can establish whether banks use the same provider. One option is to separate confirmed common causes from possible connections. The latter could carry a provisional link that can be revised, rather than being forcibly merged. This is an analytical proposal in this article, not a classification system adopted in the new agreement. It would allow effects to be communicated before the cause is established while reducing the risk that a hypothesis circulates as fact.

7. Reporting burden is more than a count of forms

For a financial institution reporting to several authorities, repeatedly entering the same number differs from rebuilding a number for another purpose. A team may know how many transactions were affected without being able to produce the unsettled value or its currency breakdown immediately. A single form leaves substantial work if different calculations are still required behind it. An exercise should examine not just the number of recipients, but who calculates each figure, from which underlying data, and which figures can serve more than one report. More shared fields do not by themselves demonstrate a lighter workload.

Alignment also requires priorities. Demanding detailed calculations during a disruption may divert staff from customer enquiries or payment checks. Deferring too much can leave authorities unable to assess spillovers to other institutions. This is not an argument that fewer reports are always better. It calls for separating information essential to immediate decisions, information that needs time to verify, and information that can be prepared in normal conditions. After an exercise, a more useful test than the number of deleted fields is whether duplicated work fell without impairing necessary decisions.

8. Separate what is needed from what may be shared

Information that a foreign authority needs for a decision may differ from information a domestic institution is permitted to send. Identifiable customer information, contractual information about counterparties and information used to respond to an attack may have different sharing conditions. This article makes no finding about any individual country’s law. The key point is that joining an exercise, or finding information useful, does not itself create permission to share it. A scenario that assumes data can be sent simply because a recipient wants it can miss a constraint that remains in a real incident.

Operationally, participants can examine alternatives to sending the entire underlying record. A design might communicate affected services and currencies, aggregate amounts and the scope of verification without naming individual customers. Aggregation, however, does not always make sharing permissible; a small set of transactions may still identify a counterparty. The test is not whether participants use the word anonymisation, but whether the information supports the recipient’s decision, the basis for sending it is established and conditions on onward sharing are understood. Explicitly identifying withheld information also makes it less likely that a recipient mistakes a blank field for no impact.

9. Communication arrangements extend beyond delivery

Providing a secure channel and making received information usable are separate tasks. An initial report might arrive over an encrypted connection yet fail to reach the decision-making team because the recipient does not know who may receive it next. Rushed forwarding in the other direction could breach confidentiality conditions. This is not an assessment of communication products. It is a question of institutional practice: recipients, acknowledgement, permitted internal users, replacement by updated versions and retention need to form a coherent procedure.

A contact list is more informative when it distinguishes roles from named individuals. Staff changes and time differences mean that reaching a registered person does not necessarily reach someone authorised to decide. In a hypothetical exercise, a contact might answer without being able to clarify an aggregate figure. Successful transmission would not resolve the problem: participants still need to know which team can explain the number and which can approve a reply to another jurisdiction. Separating acknowledgement from the ability to understand the information and choose an action helps distinguish equipment problems from organisational ones.

10. One institution or a shared service?

An institution’s size alone does not determine its relevance to the financial system. A small bank could be the first to report trouble at a service used by several institutions. The task would be to establish whether others depend on the same function, not to infer that the reporting bank has suffered a large loss. Conversely, a report from a large bank might concern an isolated part of its business. Separating questions about an individual institution’s condition from questions about shared dependencies clarifies who needs the report and what information they require.

A shared provider’s name does not establish identical effects across countries. Institutions may use different functions, contracts, connections or currencies even when they buy services from the same company. A tabletop exercise could compare decisions made by grouping reports under a provider’s name with decisions made by separating business functions. That is a proposed exercise question here, not an announced scenario. The purpose of detail is less to enrich a provider inventory than to identify which payments or customer obligations need further investigation.

11. A payment delay does not necessarily mean a cash shortfall

Payment information contains at least two different questions: whether funds are available, and whether instructions can be processed correctly and delivered. Even with an adequate account balance, a recipient may be unable to use an expected payment if its status cannot be verified. Conversely, a functioning system does not complete a payment if the necessary currency is unavailable at the necessary time. A report that merely says payments are affected leaves authorities and counterparties uncertain whether to await operational clarification or consider funding arrangements.

This distinction does not presume central-bank support. Eligibility, collateral, applications and decisions have their own conditions; the exercise agreement does not establish a new right to assistance. The analytical proposal here is to distinguish queued payments, confirmed receipts and payments whose status is unknown, even before the cause is established. Resending an instruction could create a duplicate payment, so additional funding is not necessarily the answer. Assembling information also differs from having authority to choose a response. That boundary remains during an exercise.

12. Cross-border payments face different deadlines and currencies

Even when countries receive the same incident information, the time available to act may differ. This is a general timing issue, not a comparison of actual national cut-off times. The sender may be able to wait for clarification while an approaching business deadline forces the recipient to consider alternative funding. A payment slipping into the next day could affect interest, currency hedges or payments to suppliers. Sharing when the next decision needs information, alongside the time of the next report, exposes how “we will update you later” can mean different things to different recipients.

Timestamps also change interpretation. Sending time, observation time and the reference time of an aggregate can be confused, leaving a new report with old information. A hypothetical exercise could place different reference times in otherwise identical reports and test whether recipients interpret them correctly. A common clock alone does not erase the difference between data freshness and a decision deadline. Similarly, converting an aggregate into one currency differs from possessing the currency needed for a payment. A tidy total may still be insufficient if decision-makers cannot identify where usable funds are held and in which currency.

13. Public communication serves a different purpose

Detailed exchanges between authorities cannot necessarily be released to the market. Information about vulnerabilities under response or identifiable individuals could create further problems if widely disclosed. Yet customers and businesses must still decide how to pay wages, suppliers or investment obligations while disclosure is limited. The choice need not be between releasing every confidential detail and saying nothing. Available services, the scope under investigation, actions for customers and the next update can be described within verified limits. This is a general communication principle, not a claim that the G7 agreement established a particular disclosure duty.

What markets want to know can differ from what technical teams can establish. They may be unable to give a restoration time while still being able to explain which transactions remain available. Turning an unverified estimate into a promise risks damaging trust if it fails; repeatedly saying only that an investigation continues leaves users unable to assess their exposure. An exercise could examine which parts of a public notice can change as confidential knowledge improves. The useful test is how specifically communication supports users’ next choices while protecting information that must remain restricted.

14. Separate operating costs, funding effects and credit risk

An announcement of joint AI-risk exercises does not predict financial-sector profits or market direction. Economic transmission is clearer when operating costs, funding needs and credit risk are separated. Reporting, additional reconciliation and customer support could require staff. Businesses or institutions unable to confirm receipts might seek extra cash temporarily. If payment delays persist, counterparties may also need to reassess the payer’s capacity to meet obligations. These are conditional channels, not claims that the announcement has already increased costs or funding demand.

Better reporting could help by identifying which channel is operating, rather than by preventing every loss. Treating an administrative delay as broad credit distress could be excessive. Treating persistent funding shortfalls across several payees as a communication fault could delay necessary decisions. Evaluation should therefore ask whether reports distinguished different problems and reached the appropriate teams, not just how many minutes faster they arrived. Useful decisions in an exercise and lower losses in actual markets require different evidence. Exercise results cannot simply be converted into a monetary economic benefit.

Investors should likewise avoid treating participation itself as an earnings signal. Costs may arrive before benefits, while the value of an incident avoided is difficult to isolate in company accounts. Customer departures or changed contract terms could depend on the disruption’s type, duration and communication. Actual operations, costs, contracts and customer behaviour remain necessary links between policy preparedness and company valuation. The announcement alone cannot bypass those links to justify a trading conclusion.

15. Businesses must distinguish expected receipts from usable cash

For a business, when sales receipts become usable for payments matters more than the format of an authority’s report. An importer expecting a receipt may need other cash to pay a supplier if arrival cannot be confirmed. A business with little spare liquidity could face requests to change trading terms even after a short delay. Effects would vary, however, with deadlines, currencies, alternative payment methods, contracts and cash buffers. An aggregate needed by a financial authority serves a different purpose from a company’s need to understand its own transaction.

Coordination between authorities therefore does not automatically improve guidance for businesses. Aggregates describing the overall impact need a connection to channels where individual customers can check their transactions. In this article’s analysis, an exercise should distinguish information that can be explained to customers from information restricted to authorities. Alternative payment guidance could itself create confusion unless it explains eligibility and how to avoid duplicate payments. The value would lie in clarifying what can be established under existing contracts and services, rather than adding new duties for businesses.

16. Balance representative participation with practical workload

Testing information sharing only with large institutions risks building procedures around well-staffed specialist teams. This is a design concern, not a finding about the participants in the announced exercises. At a smaller institution, one person might combine customer support, aggregation and communication with authorities. Identical information requirements can impose different time demands and leave different room for backup staff. A shared format should be examined for usability without dedicated specialists and for information that can be prepared in advance. A larger headcount alone would not show that these differences had been tested.

Not every participant necessarily needs the same volume of exercise work. Separating common contact procedures from complex cross-border reporting could allow participation to reflect each institution’s role. But simplifying an institution’s involvement should not remove a critical connection to a shared dependency. Its information-sharing role is a better basis for proportionality than size alone. The objective should be a procedure that can be maintained in ordinary operations, rather than lighter conditions designed merely to finish one exercise.

17. Turn lessons into changes that can be tested again

Finding a reporting delay is of limited use if the conclusion is merely to strengthen coordination. Repeated questions about the meaning of a number might call for clearer definitions or examples, rather than more contacts. A decision stalled by restrictions on sharing will not be fixed by changing a document’s layout. Improvements should reflect causes, identify who changes what and by when, and specify where the change will be tested. That approach is consistent with the relationship between exercises and improvement in the 2020 G7 principles; it is not evidence that a specific improvement plan has been published for the new agreement.[2]

Measurement should not focus only on apparent speed. An earlier initial report followed by major corrections is hardly an improvement if recipients decide using obsolete figures. Conversely, modestly more time at the start could clarify an essential distinction and reduce later enquiries. This article favours evaluating time to initial reporting alongside communication of material revisions and usefulness to necessary decisions. Their relative importance depends on the business function; one score can obscure the differences. None of this supplies a basis for ranking countries or institutions before actual results are available.

18. Standardisation can sometimes slow decisions

There is a strong objection to uniform reporting. Combining every authority’s request in one format could make an initial report too large and slow. Restricting it to an internationally acceptable minimum could omit information essential to a domestic decision. A format without room for exceptions may also fail to describe an event outside its categories. Standardisation is a means of helping different recipients interpret necessary information, not an end in itself. A design that prioritises filling every field could be counterproductive if it encourages unverified numbers.

The answer need not be to abandon alignment or to give domestic requests unconditional priority. Shared information could be separated from purpose-specific additions, with explicit room to report uncertainty. But lower burden should be demonstrated, for example by comparing reporting of the same event in different formats. Exercises can examine whether requested information was used, whether important questions remained and whether revisions kept pace. Finding that a common format helps in some settings while others still require tailored exchanges is not necessarily failure. Establishing the limits of its usefulness has value.

19. Three conditional paths for assessing value

In the first path, exercises improve reporting practice. Recipients’ information needs and the teams able to produce it become clearer, while initial reports and revisions are handled consistently. Even with different national powers, explaining sharing restrictions in advance could reduce misdirected enquiries. This would still not establish a fall in incident frequency or losses. Improvements would need to survive in normal procedures and remain usable after staff changes. The announcement starts this path; it does not prove the outcome.

In the second path, exercises identify institutional constraints whose resolution takes time. Practice alone cannot change legal bases for sharing or authorities’ responsibilities. In the third, the fact that an exercise took place becomes the main achievement, without demonstrable operational change. Limited public disclosure may also prevent outside readers from judging whether improvement occurred. These paths carry no probabilities or scores. Distinguishing them requires more than exercise counts: useful public detail could include the types of problems found, who owns improvements and whether they are reassessed. Absence of confidential detail is not itself evidence of no improvement.

20. What evidence would change this assessment?

The first questions for future disclosures are which AI risks the exercises cover and what roles financial authorities take. Responding to AI-enabled attacks, failures of AI used by institutions and dependencies on external services can require different information and partners. Current primary evidence does not establish that the exercises cover all of them. Any announced date or design must also be distinguished from completed results. Broad descriptions of cyber exercises should not override a narrower official scope. Undisclosed design should not be filled in with either optimistic or pessimistic assumptions.

SG Group views the agreement as an opportunity to strengthen international preparedness. The central test is whether authorities with different purposes can use necessary information within permitted boundaries, rather than whether they produce common wording. Evidence that duplicated reporting fell while important decisions remained supported would strengthen the assessment. Repeatedly unresolved interpretations or sharing bases would limit it, even if exercise counts rose. For markets and businesses, the relevant question is how far uncertainty about payments, funding and communication can be narrowed.