An AI-speed cyberattack is an attack in which generative AI automates the research, the lure and the execution, cutting the time from first access to serious harm from days to minutes. For boards, the implication is simple: an incident response that depends on convening people to make decisions will now often start after the damage is done.
The threat goes by several names: AI-enabled cyberattacks, AI-powered social engineering, industrialised phishing, machine-speed attacks and deepfake fraud. They share one feature. The human entry point, the email, the call or the video meeting that persuades someone to act, has become automated, personalised and cheap to run at scale.
Board accountability for cyber risk is already explicit. The NIST Cybersecurity Framework 2.0 (2024) added a Govern function that places cybersecurity strategy and oversight with leadership. ISO/IEC 27001:2022 requires top management to demonstrate leadership of the information security management system. In the European Union, the NIS2 Directive (EU) 2022/2555 requires management bodies of in-scope organisations to approve cybersecurity risk measures, oversee them and undertake training, and it allows them to be held liable for failures.
The speed is measurable. Booz Allen Hamilton reports that the average breakout time, the time an attacker takes to move from initial access to wider compromise, fell to under 30 minutes in 2025 (Source: Booz Allen Hamilton, When Cyberattacks Happen at AI Speed, 2026). CrowdStrike recorded an average eCrime breakout time of 48 minutes, with the fastest at 51 seconds (Source: CrowdStrike, Global Threat Report, 2025).
This analysis builds on a recent conversation with Andrew Carr, global head of threat management for Booz Allen Hamilton’s commercial incident response team.
Why do AI-speed attacks change the board’s role in incident response?
AI-speed attacks change the board’s role because the decisions that matter most now have to be made before the incident, not during it. When an attack unfolds in under an hour, there is no time to schedule a crisis call, brief the executive team and debate whether to isolate a system or stop a payment. The only decisions that will be taken in time are the ones already delegated to someone who can act at once.
This is the same lesson that applies to AI agents: a committee cannot act at the speed of an automated process. The board’s contribution shifts from reacting well to having pre-authorised the right people, with the right limits, in advance. For agents specifically, see When AI Agents Break Containment: What Boards Should Require.
Why is built-in trust now a liability?
Built-in trust is a liability because attackers no longer need to break into systems to extract money; they persuade trusted people to do it for them. Carr’s point is that the organisation’s own trust relationships are the attack surface. A familiar voice, a senior name on an email or a colleague’s face on a video call is no longer proof of identity.
The best-known example came in 2024, when an employee in engineering firm Arup’s Hong Kong office transferred around US$25 million after a video call in which the other participants, including a senior officer, were deepfakes. No system was breached. Every control that assumed a recognised person was a real person failed.
The practical fundamentals Carr highlights are not new: zero trust, data classification and least privilege. What has changed is that they are now urgent rather than aspirational, and that they apply to people and processes as much as to networks. That is why we treat zero trust as a leadership decision rather than a technology project; see Zero Trust Is a Leadership Decision. Knowing where sensitive data sits is equally a board matter, covered in Privacy and Security Are One Risk.
The AIPEX view: pre-authorise decisions, and verify people as well as systems
At AIPEX we see two board-level actions that make the largest difference, and neither is a technology purchase.
- Pre-authorise the critical decisions. Agree now who may isolate systems, suspend payments, take services offline and notify regulators, within what limits, without waiting for approval. Write it down, approve it at the right level and rehearse it. This is a delegation of authority for a crisis, and it is what lets the organisation act in minutes.
- Remove trust from high-value actions. Payments, changes to bank details, access grants and data releases above a threshold should require verification through a separate, pre-agreed channel, however senior or familiar the request appears. The board should expect this rule to apply to its own members.
How should AI be used in cyber defence?
AI should be used in defence as a layer that helps people decide faster, not as an autopilot. Carr describes its role as triaging and prioritising alerts so that decision-makers get back the time the attacker has taken from them. That is augmentation, not automation: the system narrows the choice, and an accountable person makes it. The distinction matters, and we explore it further in Automation or Augmentation?
What should the board ask the CISO, and how often should it test the answer?
The board should ask questions that test speed and authority, not just controls, and it should test the answers through an exercise at least once a year:
- If an attacker gained access at 2am, how many minutes would pass before someone with authority acted?
- Who is authorised to isolate systems or stop payments without further approval, and within what limits?
- Which of our high-value processes still rely on recognising a voice, a face or a name?
- Do we know where our most sensitive data is, and who can reach it?
- When did we last rehearse an incident with an AI-speed timeline, and what did we change as a result?
If the answers are uncertain, the gap is usually ownership rather than tooling. A Fractional Chief AI Officer can bring AI risk and cyber risk under one accountable executive. The Executive Strategy Day maps your critical decisions, who holds them and how fast they can be made. For the governance model behind pre-authorised decisions, read Who Owns the AI Agent When It Breaks?
Frequently asked questions
What is an AI-speed cyberattack?
An AI-speed cyberattack is one in which generative AI automates the research, the lure and the execution, cutting the time from first access to serious harm from days to minutes. For a full explanation, see AI-Speed Cyberattacks: What the Board Must Decide Before the Incident →
What is breakout time in cybersecurity?
Breakout time is the time an attacker takes to move from initial access to wider compromise of an organisation's systems. Reported averages fell below an hour in 2025, and the fastest cases take seconds. For a full explanation, see AI-Speed Cyberattacks: What the Board Must Decide Before the Incident →
Which cyber incident decisions should the board pre-authorise?
The board should pre-authorise who may isolate systems, suspend payments, take services offline and notify regulators, and within what limits, so those decisions can be taken in minutes without waiting for approval. For a full explanation, see Who Owns the AI Agent When It Breaks? Delegated Authority for Boards →