Bringing the Pieces Together: How THEMIS 5.0 is Building an Integrated Platform for Trustworthy AI

One of the ambitions of the THEMIS 5.0 project has always been to move beyond isolated tools and create a practical ecosystem where different trustworthiness services work together in a way that is meaningful for users. Recent ecosystem integration work, shows how this vision is becoming reality.

As the project enters its final phase, partners are focusing on integrating and validating the various components that make up the THEMIS platform. While each tool provides value on its own, the real strength of THEMIS lies in how these components connect and complement one another in our integrated platform for trustworthy AI.

A Single Entry Point for Trustworthy AI

At the heart of the platform is the Data Platform, which acts as the main gateway to THEMIS. It stores assets, enables users to upload datasets and models, and orchestrates processing pipelines while connecting users with the wider ecosystem of services. The platform has reached its target Technology Readiness Level (TRL) 5 and has already been integrated with all other components.

Importantly, the Data Platform is designed to support both technical and non-technical users, providing a unified interface for accessing assessments and navigating the wider THEMIS ecosystem.

Turning Trustworthiness into Evidence

The TAI Lifecycle Actuator (TAI-LA) provides the quantitative trustworthiness layer of the platform. Building on earlier versions, recent developments include:

• Integration with the Super Decision Engine (SDE) - customer facing agent

• Enhanced dashboards and assessment views;

• New matching algorithms for recommending methods;

• Updated fairness assessments;

• Expanded playbooks and methodological guidance.

  
  

TAI-LA produces assessment results and recommendations that help users understand where trustworthiness challenges exist and how they might be addressed. Early evidence from the pilots highlights the importance stakeholders place on having clear trustworthiness information available.

Putting People at the Centre

One of the distinctive features of THEMIS 5.0 is its emphasis on human values and perspectives. The Persona Analyser helps users interpret AI assessments through the lens of their own trustworthiness preferences and ethical viewpoints. Rather than treating trustworthiness as a purely technical concept, the component recognises that people may prioritise different aspects of AI performance. Pilot feedback has led to significant refinements, including:

• Simpler explanations of concepts such as fairness and robustness;

• More flexible answers;

• Improved terminology;

• Translation of key terms;

• The option for users to opt out of 'moral' orientation assessment.

  
  

Perhaps most importantly, the team has learned that language matters. User perceptions can be strongly influenced by terminology, reinforcing one of the project's central messages: trustworthiness is as much about people and communication as it is about technology.

Understanding Risks Beyond the AI System

Trustworthy AI cannot be separated from the wider systems in which it operates.

Spyderisk extends the scope of assessment by analysing cyber-physical risks surrounding AI systems. It considers how attacks or failures might affect trustworthiness characteristics and helps users understand vulnerabilities that go beyond the AI model itself. Integration with the wider platform is already underway, with REST and MCP endpoints implemented and tested.

Continuous Improvement Through TAIME 2.0

Another important piece of the ecosystem is TAIME 2.0, the platform's retraining engine. Designed to support continuous model improvement, it allows users to upload datasets, configure training jobs, monitor progress and manage versioned models through an intuitive interface.

The component currently supports five different systems under test, including healthcare and disinformation applications, and has achieved TRL 5. End-to-end workflows, from upload to download, have already been demonstrated.

Towards an Integrated Trustworthy AI Ecosystem

Perhaps the most significant achievement of Work Package 4 is not any individual tool, but the fact that they are increasingly working together.

The Data Platform acts as the entry point.

The Persona Analyser captures user preferences.

The TAI Lifecycle Actuator generates trustworthiness evidence and recommendations.

Spyderisk assesses broader risks.

TAIME 2.0 supports continuous model improvement.

And connecting them all is IBM's Super Decision Engine (agentic AI), which provides a more accessible and user-friendly way for doctors, journalists, port operators and other stakeholders to interact with complex trustworthiness information.

THEMIS 5.0 Ecosystem

Looking Ahead for our Integrated Platform for Trustworthy AI

With integration largely complete and validation activities accelerating, the project is entering an exciting stage. The focus now shifts to testing the platform in real-world scenarios and ensuring that trustworthy AI is not just a theoretical concept, but something that users across healthcare, ports and media can understand, assess and ultimately trust.

Because building trustworthy AI is not about creating a single tool. It is about creating an ecosystem that brings technology, people and values together.