AI Systems Face Higher Cyberattack Risk, PwC Report Finds

What happened

A new report from PwC finds that business leaders feel significantly less prepared to defend against cyberattacks targeting artificial intelligence systems than they do against traditional cybersecurity threats. The research points to a widening gap between how quickly organisations are deploying AI and how much they are investing in securing the models, data pipelines and infrastructure that underpin it.

According to Finextra's coverage of the findings, executives view AI-specific risks — such as manipulation of training data, model theft or adversarial attacks designed to corrupt outputs — as a distinct and less understood category of threat compared with conventional network or endpoint security. PwC's analysis suggests this confidence gap is a direct consequence of adoption outpacing governance: companies are rolling out AI tools faster than they are building the controls needed to protect them.

Why it matters

For organisations racing to embed AI into customer service, operations and decision-making, the report is a reminder that capability and resilience are not the same thing. An AI model that drives a chatbot, underwrites a loan decision or personalises an offer is only as trustworthy as the data and infrastructure behind it — and if security hasn't kept pace with deployment, that trust is fragile.

This matters beyond IT. Leaders responsible for digital transformation and customer experience are increasingly accountable for systems they may not fully understand from a security standpoint. A breach or manipulation of an AI system doesn't just cause downtime; it can quietly corrupt the outputs customers and employees rely on, eroding confidence in ways that are harder to detect and repair than a conventional outage.

The Renascence take

Most conversations about AI risk focus on model accuracy or bias. PwC's findings point to a quieter, more structural problem: organisations are building customer-facing AI experiences on foundations they haven't stress-tested for security.

The real exposure here isn't technical — it's behavioural. Teams move fast on AI because the upside is visible and immediate, while security investment is slow, invisible and easy to defer until something breaks. A customer-obsessed operator should treat AI security the same way it treats service reliability: as a trust commitment, not a back-office cost. That means security reviews built into every AI rollout from day one, clear ownership of model and data risk, and transparency with customers about how AI-driven decisions are protected — because the first time a model is compromised, it's the experience, not just the infrastructure, that customers will stop trusting.