KnoWare validates its prefiguration methodology and suite of tools through four real-world pilot use cases, each representing a high-impact, risk-sensitive application of General-Purpose AI. These pilots span energy and commodity trading, where a conversational assistant supports time-sensitive financial decisions; pandemic early warning, where an agentic system forecasts zoonotic disease outbreaks; maritime cybersecurity, where AI strengthens the security of critical infrastructure through real-time anomaly detection; and food fraud and chemical risk assessment, where AI reasoning is applied to detect anomalies in food safety data. Across all four cases, the project’s tools and benchmarks are deployed and tested in collaboration with the use-case partners, generating concrete evidence on how emerging GPAI capabilities and risks manifest in practice, and how they can be detected, interpreted, and governed before they translate into harm.
Four high-impact
real-world validations
KnoWare’s methodology and tools are co-designed and validated through industrial use cases in high-risk areas with significant economic potential and societal implications.
GPAI-Powered Decision Support Chatbot for Energy and Commodity Traders
This use case is focused on the energy trade sector and involves deploying a General-Purpose AI (GPAI) model as a real-time conversational assistant for traders. The chatbot is designed to integrate and analyze information from diverse sources—including satellite tracking, shipping data, market reports, and geopolitical news—to aid in decision-making under uncertainty.
The system aims to optimize trader productivity and responsiveness during high-stakes, time-sensitive decisions. However, because the system autonomously prioritizes information and recommends actions, it introduces risks of subtle behavioral shaping or misaligned incentives. Therefore, this use case provides a concrete scenario for monitoring and interpreting emergent, potentially manipulative behaviors in real-time, user-level applications.
Energy Trade
• Goal Optimisation
- ≥5 risky behaviours detected
- ≥80% explainability rate
Multimodal Agentic GPAI for Pandemic Modeling and Early Warning
Set in the health sector, this use case deploys a multimodal agentic GPAI system to predict and manage outbreaks of highly pathogenic avian influenza (HPAI), a zoonotic disease with pandemic potential. The system processes a wide array of data modalities, including climate and weather data, poultry farm records, satellite imagery, real-time wildlife reports, and social media trends, to identify early warning signs and model the severity of outbreaks.
Unlike conventional AI, this agentic system formulates its own hypotheses and dynamically adjusts its data acquisition strategies—for instance, changing its inputs if it detects a spike in wild bird deaths or sudden weather shifts. This use case allows researchers to evaluate how autonomous systems behave, accumulate risks, and make decisions in highly uncertain and safety-critical biological threat environments.
Health
• Agentic System
- ≥5 unsafe decisions identified
- ≥75% output explainability
Multi-modal GPAI framework for cybersecurity maritime event detection
Aimed at protecting critical infrastructures, this use case addresses the major cybersecurity threats of maritime spoofing and jamming, where attackers manipulate signals (like GPS, radar, and VHF) to mislead vessels and automated systems. The project will develop a multi-modal GPAI framework that fuses heterogeneous data—such as AIS, RF data, and satellite imagery—to accurately detect these deceptive cybersecurity events.
Because integrating these data sources and preventing false positives is highly complex, the backend model will be paired with a user-friendly chatbot interface. This chatbot will engage with end-users (such as maritime analysts and authorities) to explain the detected anomalies, providing necessary context and ensuring the system’s reasoning is transparent and actionable for human operators.
Critical Infrastructure
• Fine-Tuned Model
- >80% detection accuracy
- 80% events explained by chatbot
Agentic GPAI for Food Fraud Detection and Chemical Risk Assessment
This use case explores the application of an agentic multimodal GPAI system for food authenticity testing and chemical risk assessment. The system will leverage proprietary chemical analysis datasets, including near-infrared (NIR) spectroscopy and chemometric signatures, alongside supply chain metadata and textual reports, to detect food fraud and contamination.
Operating in an open-ended, exploratory fashion, the GPAI model will generate hypotheses about potential adulterants, infer fraud mechanisms, and propose new combinations of chemical tests. This scenario tests how models behave when given the autonomy to reason and adapt within a high-stakes domain, emphasizing the need to detect spurious associations or unsafe generalizations before they impact public health.
Food & Chemical Safety
• Agentic System
- ≥5 risky behaviours detected
- ≥80% explainability rate