AI Is Lowering Both the Barrier and the Time Needed for Cyberattacks: What Should Family Offices Recheck?

Three changes in attack capability, plus third-party dependency risk

The paper groups frontier AI attack capabilities into faster vulnerability discovery, easier creation or adaptation of attack tools, and the ability to connect multiple steps. Financial institutions face another layer of risk from shared cloud, software or AI providers: a single provider's outage or policy decision can propagate across institutions and jurisdictions. The two issues are related, but they are not the same risk. The paper also states that its views are those of the authors and do not necessarily represent the BIS, its member central banks or Basel standard-setting bodies.

These capabilities can shorten the time institutions have to patch vulnerabilities and respond to incidents. Attacks that previously required more people and specialist knowledge may also become possible for smaller teams. The paper also notes that AI can support defence, including anomaly analysis, software testing and faster detection; the outcome depends on how both attackers and defenders use it.

A family office's attack surface is larger than one login page

Family offices commonly spread investment, accounting, document, communication, payment and legal-administration work across internal and external systems. Identity data may sit on a file platform, payment instructions may travel by email or instant messaging, investment reports may come from banks and administrators, and external accountants or company secretaries may hold access rights.

If one provider is compromised, attackers may not need to breach the family office directly. They can use stolen email content, contacts, document formats or work patterns to create more convincing fraudulent instructions, then attempt to change beneficiary accounts, obtain one-time codes or move into other platforms.

Shared providers can amplify an individual incident

Using the same cloud, identity, software-update or AI service can improve efficiency across institutions, but it also creates common dependencies. A vulnerability in a widely used service can affect multiple banks, administrators, advisers and family offices at the same time.

Vendor due diligence should not end with a questionnaire. Families need to know where critical data sits, which subcontractors can access it, how a provider reports incidents, how accounts are disabled and whether a workable alternative process exists during an outage.

Payment and identity processes need independent verification

Fraudulent instructions can imitate familiar text, voices or images. Controls should not depend on whether an employee can tell from appearance alone that something is fake. More resilient practice is to require independent verification for beneficiary-account changes, high-value payments, asset sales, password resets and new administrator access.

The verification channel should be separate from the original instruction, such as a call to a pre-registered number or confirmation by a second authorised person. Messages that are urgent, confidential or ask staff to bypass procedure should trigger higher-level review rather than justify skipping controls.

Response time matters more than the length of the policy

During an actual incident, the family needs to decide quickly who can freeze payments, revoke access, contact banks, preserve evidence and notify affected parties. A long policy is difficult to turn into action without contact lists, offline copies and exercises.

A basic exercise can start with one concrete scenario: the main email account and cloud files are unavailable at the same time while a large payment is awaiting execution. Can the team verify the instruction and stop the transaction without relying on the affected systems?

Questions G70 would examine next

Which data, permissions and services are concentrated with one provider? Must new beneficiary accounts and high-value payments be verified through an independent channel? If cloud or communications systems fail, who can decide within one hour? Do vendor contracts specify notification deadlines and responsibility for data after an incident?

AI is making attacks faster and easier to scale. The time a family office needs to shorten is the interval between detecting an anomaly and stopping the loss.

Important notice

This article is for general information and educational purposes only. It is not cybersecurity, legal, investment or regulatory advice. Institutions should obtain professional advice based on their own systems, data and jurisdictional requirements.