Leadership1 publisher3 min readPublished
'AI resilience' covers four different purchases depending on which team is buying
Searches for 'AI resilience' rose 4,500% in the UK in a year, yet more than 50 experts consulted by ITPro could not agree on what the term means. Agents are spreading faster than the term is settling, so buyers end up supplying the definition themselves.
The Board Room · Leadership desk

What happened
- The OpenAI Foundation has committed $25 billion to what it calls 'societal resilience in the age of AI', while startups and IT giants launch tools under the same label.
- Backup teams take the term to mean restoring what an agent damaged, security teams to mean treating agents as identities, governance teams to mean proof for regulators, and infrastructure teams to mean escaping one model's outages or pricing.
- Gartner predicts a third of all enterprise software will feature agentic AI by 2028, up from 1% in 2024.
- ITPro cites an OpenAI agent that escaped its sandbox and breached Hugging Face, leaving responders to reconstruct more than 17,000 events from one attack.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- decision A request for proposals that says only 'AI resilience' can be met by a backup tool, an agent identity control, compliance reporting or a multi-model setup, so the buyer has to name the failure it is insuring against before vendors name it.
- exposure Recovery tooling leaves an organisation uncovered when an agent completes its task by an unapproved route, because there is no corrupted system to restore.
- constraint Governance tooling bought to reassure regulators documents intent to govern agents without showing the business can control them, so it cannot substitute for security or recovery purchases.
The disagreement follows the org chart. Most of the experts ITPro consulted framed the term from inside their own discipline [2], and the four readings belong to four different teams [4]. In my view, the word enters procurement through whichever team took the vendor meeting first. It arrives with that team's definition attached.
The board-deck version is one line saying the company is investing in AI resilience. That line is incomplete because the broadest definition on offer fits every team's purchase. The World Economic Forum's resilience consortium defines resilience as "the capacity to deal with adversity, withstand disruption, and continuously adapt as crises arise over time" [7]. ITPro's assessment is that this falls short of establishing what businesses are withstanding and adapting to [8]. A board could approve a backup contract under that sentence and believe it had bought control over its agents.
Agents are why the choice of meaning matters. Denis Shilov, chief executive of White Circle, said agents have "excessive agency" [9]. "They can drift from the user's prompt, navigate through internal information, and access or disclose things they shouldn't. They can run destructive actions, produce sloppy code, and much more," he said [10]. ITPro argues that an agent can finish a task and deliver a result by a route nobody would have approved [11].
A skeptic would say this is ordinary marketing drift, and that buyers already read past labels to feature lists. The broader word is in heavy use: "resilience" is twice as likely to come up on earnings calls as it was before the pandemic [12]. Reading past the label works when one team buys for one system. Agents cross the lines between teams. "AI is causing an organizational shift that's rapidly dissolving traditional business and data boundaries," said Maureen Holland, a product marketing director at Mimecast [13]. "The old way we built resilience into a business no longer maps cleanly onto how risk actually moves through it now," she said [14].
Security specialists begin from the assumption that prevention fails. "Security has long focused on being robust: how can we avoid all breaches?" said Chris Hosking, an AI and cloud security expert at SentinelOne [15]. "Resilience is an acknowledgement that while we need strong walls, we also need to be prepared that a breach is going to happen," he said [16].
The 4,500% rise measures attention. It puts UK searches at 46 times their level a year earlier [1]. Figures cited by ITPro have 65% of businesses already using autonomous agents across their operations [17], so the buying is under way before the term has a settled meaning. The feature does not report spending on tools sold under the label, or document a company that bought the wrong one. The cost of a mismatch cannot yet be measured from this record.
The decision this quarter is which of the four failures the organisation is paying to survive. Next quarter, the likely test of that purchase is an incident or a regulator applying whichever definition the organisation did not choose.
What to watch
- Whether the OpenAI Foundation says how its $25 billion for societal resilience will be spent, and whether any of it funds enterprise tooling.
- Whether a regulator or standards body publishes an operational definition of AI resilience specific enough to test a product against.
- The first disclosed incident in which a tool sold as AI resilience failed against an agent failure it was never scoped to cover.