Ten solutions from Zurich’s 2026 Agentic AI Hyper Challenge have secured sponsor backing to move towards production. Their progress shows how employee-led innovation can turn emerging technology into practical solutions for customers and the business.
More than 700 Zurich employees across 24 countries took part in the 2026 Agentic AI Hyper Challenge, which this year was focused on participation internally. In just a couple of weeks, they built over 200 prototypes and more than 1 million lines of code, consuming over 200 trillion tokens. And not all the teams were techies; in fact, most didn’t have formal experience developing, which demonstrates how accessible AI-enabled development is becoming.
The scale was impressive. What happens next matters even more.
Ten solutions have been greenlit by business sponsors to move towards production. They address priorities across underwriting, claims and customer experience, spanning Commercial Insurance, Specialties, Life and Retail.
The teams began with problems they knew firsthand. Some were highly technical, such as strengthening security testing. Others focused on familiar operational challenges, including reviewing large volumes of documents, detecting potential fraud or finding relevant information from previous insurance submissions.
The result is a varied set of solutions rooted in the realities of Zurich’s business.
Exploring new AI capabilities
Certain solutions are research-led. They investigate emerging AI capabilities and new ways of working while helping Zurich understand their potential, limitations, and risks.
The Autonomous Red Teaming solution, developed by Token Boosters, focuses on cybersecurity. Traditional penetration tests provide a view of potential weaknesses at a particular moment in time. This solution uses attacker and defender AI agents to test systems continuously, identify possible routes to sensitive data and automatically defend against them.
For Zurich, the potential value lies in identifying vulnerabilities earlier and giving teams clearer information on where to focus their efforts and in this fast-moving space aspects of how Token Boosters built their solution are already included into Zurich’s live cyber defenses.
What surprised our Information Security team is how far teams without security or hacking experience could progress on this use case and how it lowers the bar for malicious actors.
NeuralNomads took on a different challenge: the volume and complexity of information handled by teams in the Life business unit. Correspondence, calls and other documents need to be reviewed, classified and directed to the right place. This can take time, and important signals may not always be easy to identify.
The team used a methodology developed by AI researcher Andrej Karpathy to identify the prompts most likely to classify correspondence accurately. They then went beyond the original use case, creating an Auto Research platform that could potentially be applied to optimize prompts for other AI solutions. The platform includes confidence-based promotion, rollback mechanisms, audit and reasoning trails, and human oversight.
The benefit is practical: less manual triage, greater consistency and earlier identification of cases that may involve vulnerability, regulatory concerns or require particular attention. Their platform is already being used in three countries accelerating how we can achieve the right accuracy in our AI solutions.
The third research-led solution considers how customers may interact with insurers as personal AI assistants become more widely used. Developed by the CommPAEnion team, it enables Zurich AI agents to help customers, and their AI agents to understand, quote and purchase DA Direkt pet insurance.
The solution includes guided workflows, policy-based answers, and consent controls. It explores how Zurich could support agent-to-agent interactions while keeping customers informed and in control of decisions made on their behalf.
Addressing business priorities
Other solutions have a more immediate focus on business and customer outcomes.
Pricing work is resource-intensive: reconciling data, stress testing models, interrogating portfolios, iterating. NEON's AI pricing agents work across the whole chain — cleaning and enriching data, exploring portfolios, developing pricing models, and rating individual risks.
Actuaries and underwriters delegate to the agents, which show their working at every step. Zurich's experts can then focus their experience where it matters most: framing the right questions, judgement on key assumptions, and turning analysis into decisions the business can act on.
TruthLens addresses fraud detection across documents, images and videos. These checks can involve large amounts of information from several sources. Investigators need to identify connections and warning signs while retaining clear evidence for review and decision-making, many of these are invisible to the human eye and deep faking documents becomes easier and more frequent.
Its deep fake detection solution uses specialist agents to perform cross-file checks and even a pixel level analysis of the documents then follows up on risk signals and produce explainable fraud assessments with an audit trail. The intention is to support more consistent evidence gathering and help investigators concentrate on the cases requiring deeper examination while allowing the claims with genuine documentation to be resolved quickly for our customers.
BBX and CanAIda, meanwhile, explored how historical information could support cyber underwriting. Zurich may receive the same risk as new business more than once, but information from previous submissions is not always readily available or easy to compare.
Their solution combines current and past submissions into a clearer risk history. It standardizes information, highlights changes, and provides relevant comparisons, giving underwriters a stronger evidence base for evaluating repeat risks. By applying AI to that historical record, the solution could help underwriters assess repeat risks more objectively and consistently and give them a stronger evidence base for negotiations.
Lastly, ZRoot focused on supporting Zurich’s Mid-Market growth ambitions. The team developed an AI-powered Signal Intelligence solution that uses multiple AI agents to analyze internal and publicly available external data, identifying businesses that closely align with Zurich’s risk appetite.
By surfacing high-potential prospects earlier in the sales cycle, the solution could help market-facing teams focus on opportunities with a greater likelihood of underwriting success and profitable growth.
These solutions are very different, but each begins with a clearly defined opportunity and a practical view of where AI might help. That grounding will be critical as the teams move into the next stage.
Developing Zurich’s AI talent
The Hyper Challenge has also revealed capability within Zurich.
Proof that you don't need a cybersecurity background to win a cybersecurity challenge — you need to learn how to think together with an AI. That's what this challenge really taught me.
Alessandro Paradiso
PropTech Experience Owner
Hundreds of employees developed their AI skills through the program, and more than 60 high potential internal AI talents were identified. Zurich plans to support their continued development and draw on their experience to address further AI opportunities across the Group.
The hackathon provided an incredible opportunity to deepen my AI knowledge while collaborating with colleagues across the globe. It reinforced that innovation happens when continuous learning, teamwork, and technology come together to create meaningful business impact.
Supriya Subbarao
IT Advisory Enterprise Architect
This gives the program value beyond the individual prototypes. Participants learned by building, testing and working across disciplines. Business experts brought their understanding of customers, products and processes, while technologists helped teams explore what emerging AI tools could achieve.
I’ve led many AI initiatives from the business side, but the Hyper Challenge reminded me how different it is to move from discussing what technology can do to actually building with it. Getting hands-on deepened my understanding, challenged some of my assumptions, and gave me new perspectives to bring back to my colleagues.
Mattia Visigalli
Innovation Consultant
That combination will be important as Zurich develops and applies AI more broadly.
The Hyper Challenge allowed me and three other colleagues from different departments in Australia to come together and work on an exciting new cutting edge area of AI that we were interested in. This challenge gave us the opportunity to learn, experiment, and create something meaningful that could deliver real value to the business. Seeing our solution already being used around the Group is really the icing on the cake.
Madeline Tourne
Transformation Lead - Claims Ops & Change
What happens next
Sponsor backing is an important step, but it does not guarantee that every concept will reach production in its current form.
The ten solutions will need to be tested, refined, and validated. Teams will assess how well they perform in real business settings, how they connect with existing processes, and where human oversight is required. Security, governance, and responsible AI considerations will remain part of that work. And of course, old fashioned change management is always needed to ensure the change sticks, and the value is delivered 30 years on, and especially in the rapidly moving world of AI John Kotter’s findings continue to apply.
Beyond these winning solutions and the talents developed, the Hyper Challenge has been an opportunity for Zurich to learn about AI can best add value to the Group in a very concentrated window. It has identified many business and technical learnings about how to build, test and govern AI systems which we will apply in our future work to accelerate how we deliver value to our customers and stakeholders.
