Half of respondents to PwC’s newly released 2027 Global Digital Trust Insights survey identified attacks on AI systems as a threat their organisations were least prepared to tackle over the next 12 months. The findings place AI security ahead of cloud threats, third-party breaches and ransomware in businesses’ assessments of their readiness, according to reporting by Risk Management Magazine.
The global survey points to a tension for organisations adopting the technology: AI is both their leading cybersecurity investment priority and an area of substantial perceived vulnerability. For UK and European resilience leaders, the findings also raise questions about dependencies on external models, data and platforms that can put critical operations beyond an organisation’s direct control.
What happened
Respondents identified several distinct forms of AI-enabled attack they expected their organisations to need to address. Autonomous botnet compromise was cited by 53 per cent, while adversarial attacks and data poisoning each attracted 52 per cent. Malware capable of adapting or changing its form to evade detection followed at 46 per cent; deepfake audio or video used for fraud or impersonation was cited by 44 per cent.
Those categories pose different defensive challenges. Data poisoning can undermine an AI system through the information used to develop or operate it, whereas deepfake impersonation targets confidence in a person’s identity. Treating both as a single technology risk could obscure the separate responsibilities for data integrity, access controls and transaction verification.
Beyond AI, 40 per cent of respondents highlighted cloud-related threats among their readiness concerns. Third-party breaches stood at 34 per cent, with ransomware and software supply-chain compromise each at 33 per cent. These are measures of perceived preparedness, not a forecast of how frequently each type of attack will occur.
Spending priorities reflect that mix of emerging and established risks. AI investment was a cybersecurity budget priority for 58 per cent of respondents, followed by cloud security at 46 per cent. Data protection and trust accounted for 39 per cent, network security and zero trust for 35 per cent, and threat management for 33 per cent. The figures identify investment priorities rather than the proportion of budgets allocated to each area.
PwC’s assessment is that security and oversight must develop alongside AI adoption, rather than follow deployment. That puts investment decisions within a wider accountability debate, also explored in our coverage of calls for CISOs to make AI risk a boardroom decision. Buying capabilities does not, by itself, establish who can authorise their use or accept the associated operational risk.
The background
Geopolitical uncertainty is another force reshaping cyber planning. PwC describes cyber operations as an enduring instrument of competition between states, during both conflict and peacetime. It expects that environment to influence corporate cyber strategy over the coming year, including how organisations select suppliers and secure software development.
Half of respondents were changing vendor, third-party or supply-chain risk management in response. A further 49 per cent reported changes covering cyber insurance, incident response and crisis management planning. Threat intelligence, geopolitical monitoring and collaboration with government or industry featured for 47 per cent; changes to critical infrastructure location, resilience and redundancy were cited by 43 per cent.
Supplier diversification can reduce dependence on one provider, but it also creates relationships and interfaces that require scrutiny. PwC warns that evolving technology networks can leave security gaps. Our reporting on the Engie customer data breach and partner marketing systems examines the related issue of exposure through business partners.
AI adds another layer to that dependency problem because organisations may rely on externally supplied models, platforms, services and data simultaneously. PwC also identifies continuing weaknesses in data protection as businesses consider granting AI agents greater autonomy. The oversight challenge extends to the defensive tools themselves, a subject examined in our coverage of governance gaps in AI-built security tools.
What people are saying
PwC’s recommendations reach beyond security departments to the executives responsible for finance, operations and corporate risk. As reported by Risk Management Magazine, the consultancy identifies a gap between discussing cyber threats at senior level and making them a routine consideration in business decisions. Executive attention is not the same as an effective decision-making process.
The report assigns roles to chief executives, chief information security officers, chief risk officers, chief financial officers and chief data officers. Their task is to provide information that allows commercial priorities and cyber exposure to be weighed together. That requires more than technical reporting: leaders need to understand which business activities could be affected and what choices are available.
On continuity, PwC urges organisations to identify the minimum capabilities necessary to remain operational during disruption, then develop and test recovery arrangements around them. This is a narrower and more demanding starting point than attempting to restore every system at once. It makes the preservation of essential services, rather than the recovery of technology alone, the organising principle.
For continuity teams, the practical consequence is that recovery priorities cannot be settled solely by IT. Operational leaders must establish what the business can temporarily do without, which functions cannot stop and how reduced services would work. Exercises can then test whether the technical recovery sequence supports those requirements, rather than merely demonstrating that individual systems can be brought back online.
PwC also argues that technology modernisation should be assessed for both commercial benefit and security. That changes the investment question from whether a tool offers useful capabilities to whether those capabilities can be operated reliably. A project’s business case is incomplete if the resources needed to protect and sustain its operation sit outside the assessment.
Its final recommendation concerns the working relationship between employees and AI agents. The consultancy calls for workforce development and explicit human responsibility as automated systems take on more tasks. The implication is organisational as much as technical: staff need the skills and authority to supervise agent activity, not simply access to new tools.
What happens next
The report sets out a programme of work rather than evidence that these changes have been completed. Its recommendations should therefore be read separately from respondents’ reported priorities. An intention to strengthen resilience does not establish that recovery arrangements have been tested or that responsibility for an automated workflow has been assigned.
One issue to watch is whether organisations translate their continuity objectives into demonstrable operating capability. A recovery exercise can expose a mismatch between what executives expect to remain available and what teams can actually sustain. For UK and European businesses, that distinction is important wherever customer services depend on several departments recovering in a coordinated sequence.
Another is technology concentration. PwC asks organisations to understand where dependencies could magnify disruption. The relevant question is not simply how many suppliers a business uses, but whether apparently separate services share an underlying provider or platform. That analysis can reveal why the loss of one component might disable several functions simultaneously.
As human-and-agent working develops, governance will also need to become specific enough to guide daily operations. The practical questions include who reviews an agent’s output, who can intervene when its behaviour is unexpected and who remains responsible for the resulting business action. These are implications of PwC’s proposed operating model, not controls the survey establishes are already in place.
Workforce preparation will be a further test of implementation. Training needs to equip employees to recognise when automation requires scrutiny and to escalate concerns through an understood process. Without that operational detail, a named human owner may carry responsibility without having the knowledge or means to exercise meaningful oversight.
