How Zurich US built AI-ready resilience across a multi-cloud enterprise
As its AI capabilities expanded, the organisation sought a new level of reversibility
Insurance shopping isn’t the chore it once was. These days, people can receive a quote almost instantaneously. But behind that convenience lies a complex process: insurers pulling together data from public records, driving histories, vehicle details and more—a massive co-ordination effort across scattered data sources.
It’s exactly the sort of problem AI is built to solve, a fact not lost on Zurich US. The provider of insurance and risk management services is using agentic AI across everything from fraud detection in claims and software engineering to customer and broker experience. Some of its most significant impact has been in risk assessment. Each year, Zurich’s underwriting teams process hundreds of thousands of submissions. AI-powered agents rapidly surface actionable insights, relevant data and key risk indicators for every submission, enabling underwriters to make faster, more informed and comprehensive decisions. “We don’t view AI as a standalone initiative,” says Madhu Ramamurthy, chief information officer at Zurich US. “We see it as a capability embedded within our business processes to help our teams work more effectively and deliver better outcomes.”
But as agentic AI helps insurers unlock new opportunities, it also raises the stakes on security. In response, Zurich has rethought its approach to cyber, partnering with data-security and AI-operations company Rubrik to tackle resilience challenges.
The new resilience imperative in the age of AI
As organisations push to deploy AI agents and unlock productivity gains, a new risk environment is unfolding that makes security incidents a matter of when, not if. Some 98% of organisations with operational agents have already experienced an incident, according to a recent Economist Enterprise white paper supported by Rubrik. Agents need a level of freedom to deliver the greatest value, but granting it opens the door to risk
Any increase in IT complexity—an inherent byproduct of agentic AI—creates fresh cyber-vulnerabilities and entry points. “A lot still relies on the underlying foundation,” says Adam Page, Zurich's chief information-security officer for Europe and the Middle East. But adaptation is critical, he says, because agentic AI introduces entirely new attack vectors.
One of the most significant vulnerabilities comes in the form of prompt injection, in which attackers feed inputs into a large language model (LLM) that causes it to act maliciously. A 2025 Nature Communications study testing 594 prompt-injection attacks found that all evaluated AI models were vulnerable to manipulation.1 Other risks include data poisoning, in which attackers tamper with the data used to train an LLM, and sensitive information disclosure, where models inadvertently expose confidential information.
Even more challenging, AI allows attackers to chain together vulnerabilities in a way they couldn’t before. “Individually, they may not have made much impact,” says Mr Page. “Together, they do.”
Individually, [vulnerabilities] may not have made much impact. Together, they do."
Adam Page, Zurich's chief information-security officer for Europe and the Middle East
Building modular reversibility with Rubrik
Alongside these new AI risks, Zurich US has also been managing growing IT complexity. Over the last 18 months, the company migrated 160 applications from its data centres into a multi-cloud environment. “We wanted to further strengthen our resilience,” says Mr Page, who was chief information-security officer for the US at the time. “That’s when we put in Rubrik.”
In the agentic age, security is not just about prevention. It’s also about reversibility: the ability to recover quickly and precisely while minimising disruption to the business. Rubrik gave Zurich a centralised view of its backups, including their health and any anomalies, while also helping the company rigorously test its resilience capabilities.
The organisation can now take a more modular approach to recovery across its infrastructure, which still includes some on-premises storage alongside its multi-cloud ecosystem. Disaster recovery used to be largely all-or-nothing, Mr Page says. With Rubrik, Zurich can recover specific systems or application environments independently. If the claims system were compromised, for example, the company could restore it without affecting applications elsewhere in the business.
As the threat landscape evolves, Zurich US and Rubrik have developed a close working relationship. “They bring their thoughts and strategy to us, and we put an operational lens on it,” Mr Page says.
Building AI-ready security
Today, Zurich is better positioned for resilience.
What has changed most, according to Mr Page, is how quickly the company can detect a problem and begin recovery with confidence. Immutable, pre-validated recovery points eliminate much of the time previously spent determining where to roll systems back to, while AI-driven anomaly detection flags suspicious behaviour in real time when indicators of compromise are triggered.
Rubrik’s built-in AI assistant, Ruby, also continues learning over time, helping Zurich identify application-specific recovery patterns and accelerate restoration further.
But Zurich executives say that the broader impact goes beyond operational efficiency. By reducing administrative work, AI is giving underwriters more time to focus on customers and higher-value activities. “It allows our teams to spend more time identifying the right risks, writing profitable business and strengthening customer relationships,” says Mr Ramamurthy. “That also creates more opportunities to better serve customers across a broader range of products and solutions.”
It allows our teams to spend more time identifying the right risks, writing profitable business and strengthening customer relationships. That also creates more opportunities to better serve customers across a broader range of products and solutions."
Madhu Ramamurthy, chief information officer at Zurich
Still, Zurich believes it has only begun to tap agentic AI’s potential. Mr Ramamurthy advocates a measured approach to deployment, starting with targeted use cases and ensuring governance keeps pace with innovation. As Zurich continues rolling out AI agents, he expects advances in predictive modelling and real-time risk assessment, as well as more personalised customer experiences.
“We’re scaling it across the organisation and identifying additional opportunities as adoption grows,” he says. “At the same time, AI tools are enabling faster, higher-quality and more cost-effective technology delivery.”