Core Project Objectives
KnoWare aims to spot new abilities in general-purpose AI (GPAI) models before they fully appear. Today’s AI evaluations mostly measure what a model can already do. KnoWare looks for the early signals of what it may soon be able to do, and for the risks that come with it. The project has four objectives.
- Develop a new scientific framework that helps GPAI providers, the EU AI Office and organisations forecast and understand how these capabilities emerge.
- Build a suite of decision-support tools and an open benchmark to catch emerging capabilities early and explain the risks they carry.
- Study what these capabilities mean for organisations and for society, including how people’s own skills and behaviour change as they work alongside AI.
- Give legal guidance on assessing and reducing systemic risks under the EU AI Act. KnoWare will test these results in four real-world pilots: energy trading, pandemic early warning, maritime cybersecurity and food fraud detection.
The goal is safer, more trustworthy AI in Europe. Understand. Anticipate. Govern. The KnoWare way.
Scientific Level
To develop a novel methodological framework for assessing and validating emerging capabilities and associated risks. This framework enables General Purpose AI (GPAI) providers and oversight bodies to forecast, interpret, and understand the complex factors and interrelationships within GPAI systems.
- Faster decisions with reliable data
- Reduced operational costs through automation
- Scalable systems that grow with your needs
- Higher ROI from every initiative
Technological Level
To design and develop a suite of open-source decision-support tools and novel benchmark tests. These are intended to improve early prediction, explanation, and interpretation of emerging capabilities and manage (systemic) risks across a model’s lifecycle.
- Faster decisions with reliable data
- Reduced operational costs through automation
- Scalable systems that grow with your needs
- Higher ROI from every initiative
Societal and Business Level
To investigate and evaluate the socioeconomic and sociotechnical implications of forecasted emerging capabilities for organizations and society. This includes monitoring the co-evolution of human and GPAI capabilities and assessing environmental impacts like energy consumption.
- Faster decisions with reliable data
- Reduced operational costs through automation
- Scalable systems that grow with your needs
- Higher ROI from every initiative
Legal Level
To develop legal guidance and roadmaps for the assessment and mitigation of systemic risks posed by GPAI models. This objective focuses on providing actionable recommendations for compliance with the AI Act and supporting the AI Office in its enforcement duties.
- Faster decisions with reliable data
- Reduced operational costs through automation
- Scalable systems that grow with your needs
- Higher ROI from every initiative
KnoWare Solutions
Key Exploitable
Results & Tools
The KnoWare project delivers a comprehensive portfolio of eight Key Exploitable Results (KERs) designed to shift the evaluation of General-Purpose AI (GPAI) systems from reactive observation to proactive anticipation and governance. To address the rapid and unpredictable emergence of new GPAI capabilities, these KERs provide a cohesive set of novel conceptual frameworks, open-source decision-support tools, and dynamic benchmarking suites. Beyond technical evaluations, the project’s results also feature evidence-based methodologies to thoroughly assess the socioeconomic, sociotechnical, and environmental impacts of AI capabilities. By integrating these practical, open-source instruments with targeted regulatory roadmaps and guidelines, the KnoWare KERs equip model providers, deployers, and policymakers with the actionable resources needed to identify and mitigate systemic risks, ultimately ensuring the safe, trustworthy, and human-centric deployment of AI in full compliance with the EU AI Act.”
Capabilities Roadmap and Capability-to-Risk Mapping for GPAI Systems
This is a structured roadmap that links General-Purpose AI (GPAI) capabilities to systemic risks. It will enable model providers, developers, researchers, and regulators to anticipate and prevent harmful outcomes.
GPAI capabilities prefiguration methodological framework
A model-agnostic framework designed to assess both latent and observed capabilities in GPAI systems . It ensures that evaluations remain consistent, transparent, and aligned with regulatory needs.
KnoWare Open-Source Suite of tools
A modular and interoperable suite of tools intended for real-world GPAI testing . It equips developers and deployers with practical instruments to ensure transparency, trustworthiness, and compliance.
GPAI Benchmark Suite and Evaluation Protocols
Harmonised benchmarks and evaluation protocols that are validated in real-world pilot scenarios. This aims to establish a shared European reference for testing GPAI performance and safety.
Emerging GPAI capabilities socioeconomic and sociotechnical impact assessment methodology
Evidence-based methodologies created to assess the societal and economic implications of emerging GPAI capabilities and their associated risks, supporting inclusive and responsible governance.
Emerging GPAI capabilities environmental impact assessment methodology
Tailored methodologies designed to evaluate the environmental footprint of the capabilities of GPAI systems, connecting the growth of these capabilities with sustainability metrics.
Regulatory Impact Analysis Framework for GPAI Systemic Risk Management
Regulatory guidelines that integrate technical evidence into systemic risk management, directly supporting enforcement of the AI Act by relevant authorities.
Regulatory Alignment Roadmap and Guidelines for Regulated Entities
Practical, step-by-step guidelines that translate regulatory requirements into actionable compliance pathways for GPAI developers, deployers, and organisations.