G7 agrees to hold exercises to prepare for AI cyberattacks
Financial authorities will rehearse their response as Britain strengthens preparations for disruption.
G7 finance leaders have approved exercises to prepare for cyberattacks enabled by artificial intelligence, Japan’s finance minister said on Thursday. The agreement concerns preparations for possible disruption, rather than an announcement that an attack has occurred.
Satsuki Katayama announced the agreement after finance ministers and central bank governors met in Bangkok on 15 October. According to Reuters, a Japanese finance ministry official said the simulations would test response procedures and inform revisions to incident manuals. Private financial institutions could participate, depending on the scenarios chosen. [Reuters](https://www.investing.com/news/economy-news/g7-to-launch-exercises-to-guard-against-ai-cyber-attacks-4953109)
The announcement leaves practical details unresolved. The Reuters report does not provide an exercise date, a list of participating banks or the scenarios authorities will use. It therefore confirms ministerial agreement to conduct rehearsals, but does not establish when individual institutions will take part or what their involvement will require. [Reuters](https://www.investing.com/news/economy-news/g7-to-launch-exercises-to-guard-against-ai-cyber-attacks-4953109)
For Britain, the agreement follows a separate domestic rehearsal. On 8 October, the Cross-Market Operational Resilience Group held SIMEX26, testing a global disruption to cloud services. The Bank of England said 38 major banks and financial-market infrastructure operators participated alongside financial authorities. That exercise concerned continuity of financial services during a simulated outage. [Bank of England](https://www.bankofengland.co.uk/news/2026/october/cmorg-holds-sector-wide-cloud-outage-simulation-exercise)
The Bank said firms should act on the lessons from that rehearsal to strengthen their ability to respond to severe disruption. It also explicitly stated that the scenario was not based on specific threat information. The distinction matters: a preparedness exercise tests an institution’s response to a hypothetical event and does not, by itself, reveal an imminent threat. [Bank of England](https://www.bankofengland.co.uk/news/2026/october/cmorg-holds-sector-wide-cloud-outage-simulation-exercise)
International coordination also predates Thursday’s decision. A US Treasury statement published on 31 July recorded a G7 cross-border exercise completed on 18 May. It brought together finance ministries, central banks, banking supervisors and market authorities around a simulated large cyberattack across G7 jurisdictions. The statement also described an agreed strategy for more frequent and consistent exercises. Those earlier activities provide background to the newly announced AI preparations. [US Treasury](https://home.treasury.gov/news/press-releases/sb0583)
The Bank’s July financial stability report explains the concern behind this work. It assessed that increasingly capable AI could help identify and exploit software weaknesses more rapidly, while also offering tools to improve defence. Shared technology suppliers and infrastructure create a further challenge: disruption to a common dependency could affect several financial institutions at once, rather than remain confined to one firm. [Bank of England](https://www.bankofengland.co.uk/financial-stability-report/2026/july-2026)
That assessment came with an important qualification. The Bank said controlled AI cyber tests without active defenders did not demonstrate reliable, undetected autonomous attacks against well-defended real organisations. Its policy concern was the speed of change and whether firms could adapt their protection, response and recovery arrangements. For households and businesses, the practical issue is the resilience of everyday services such as payments, saving and borrowing. [Bank of England](https://www.bankofengland.co.uk/financial-stability-report/2026/july-2026)
Fact-check Approved · Priya Shah — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The announcement falls within the 48-hour window and is attributed to the minister through Reuters. Official documents corroborate the background, with earlier exercises clearly distinguished from the new agreement.
- confirmed Katayama announced ministerial approval after the 15 October Bangkok meeting. — Reuters dates her remarks to Thursday. source
- confirmed Exercises will test procedures; private-sector participation is conditional. — Attributed to a Japanese ministry official; no timetable is supplied. source
- confirmed SIMEX26 involved 38 major financial firms and infrastructure operators on 8 October. — The Bank’s 9 October release specifies the date, participation and cloud-disruption scenario. source
- confirmed SIMEX26 was not based on specific threat information, and firms were expected to act on its lessons. — Both points appear explicitly in the release. source
- confirmed A previous G7 cross-border exercise concluded on 18 May, with a longer-term exercise strategy adopted. — The Treasury’s 31 July statement describes participating authorities and the simulated multinational attack. source
- confirmed The Bank assessed AI as a potential aid to attackers and defenders, with shared dependencies creating wider risks. — These are attributed risk assessments, not claims that the projected disruption has occurred. source
- confirmed Controlled tests did not establish reliable autonomous attacks against well-defended real targets. — The report explicitly distinguishes tests without active defenders from real-world capability. source
Editor's note
The new agreement is supported by Reuters’ reporting of named ministerial remarks on 15 October, published on 16 October. Primary documents substantiate the separately dated background; a primary announcement specifying the new exercises’ timetable was not located. No evidenced political disagreement over this agreement was found, so a three-way political division would be artificial.Sources
- G7 to launch exercises to guard against AI cyber attacks — Reuters, published by Investing.com
- CMORG holds sector-wide cloud outage simulation exercise — Bank of England
- G7 2026 Cross Border Coordination Exercise — US Department of the Treasury
- Financial Stability Report — July 2026 — Bank of England
More on this in other media
- Channel 4 News — „banking artificial intelligence”
- BBC News — „banking artificial intelligence”
- The Guardian — „banking artificial intelligence”