Developed by Datrix, AIA Guard is a pioneering solution in Europe for machine learning model security, designed to address risks related to AI cybersecurity.
Artificial intelligence is becoming an essential tool for businesses, revolutionizing sectors such as healthcare, finance, and digital services. However, as applications increase, so does the risk of vulnerabilities and targeted attacks. To address this challenge, Datrix has developed AIA Guard, a cutting-edge platform that identifies weaknesses in machine learning systems, ensuring security and GDPR compliance.
Now, thanks to STAR4AI, AIA Guard is evolving with new modules that expand its capabilities and open new horizons in the ethical and transparent management of artificial intelligence.
The project was carried out thanks to the contribution of the Datrix Group’s internal R&D team, led by Matteo Bregonzio, CTO and Head of R&D Datrix Group, in collaboration with the University of Tuscia and the Cyber 4.0 Competence Center.
The work team, composed of young talents with an international scope, consists of Yuliia Sobko – Project Manager, Elia Broggio and Alsu Shakirzyanova – Data Scientist and Data Engineer, and Giovanni Di Noia – Backend Developer. We spoke with them to better understand how this platform works.
AIA Guard: a PIONEERING Cybersecurity platform
Developed by Datrix, AIA Guard is a pioneering solution in Europe for machine learning model security, designed to address risks related to AI cybersecurity. By automatically analyzing a model’s entire workflow, the platform identifies and mitigates threats such as data poisoning, model theft, inference attacks, and adversarial machine learning. AIA Guard also integrates advanced features for data sanitization, protecting sensitive information and ensuring GDPR compliance.
The SATML-B project represents a further step forward, aiming to bring AIA Guard from TRL 6 to TRL 8 (Technology Readiness Level), meaning from a solution in the industrial validation phase to a product ready for commercialization. Thanks to SATML-B, the platform is enriched with tools to improve model reliability, protect data, and ensure security even in complex scenarios. Among the main innovations introduced are advanced stress tests to evaluate model resilience, solutions to mitigate adversarial attacks, and features for the anonymization of sensitive data.
Bregonzio comments: “The distinguishing element lies in its FLEXIBILITY which can in fact be implemented locally or on private clouds, without the need for interactions with external sources, making it ideal both for highly regulated environments and for organizations that require autonomy in data management. The platform was also designed to be intuitive and accessible, both for expert technicians and less specialized users. Its capabilities include scanning source code to detect known vulnerabilities, identifying sensitive data leaks, and assessing risks related to malicious inputs, such as evasion attacks that could hijack models towards unintended behaviors.”
Through penetration tests and detailed analysis, AIA Guard provides clear reports with operational recommendations, helping companies not only mitigate risks but also improve the efficiency and resilience of their artificial intelligence systems. The platform positions itself as a concrete response to the growing need for security in the public and private sectors, demonstrating STRONG COMMERCIAL POTENTIAL in a context where AI adoption continues to grow rapidly.
STAR4AI: an innovative, now mature technology aimed at the market
With the STAR4AI project, Datrix expands the scope of AIA Guard, aiming to reach level TRL 8, a milestone that represents mature technology, ready for integration into real operational environments. This is particularly significant in a context where the European AI regulation – AI Act – requires increasing transparency and reliability in the use of technologies based on artificial intelligence.
Alsu explains further: “TRL is a scale that measures the degree of technological maturity of a product or process, from 1 (basic research) to 9 (technology ready for operational use). TRL level 8 indicates that the system has been fully developed, tested in real environments, and qualified for final implementation or for integration into a production line.
In the context of the STAR4AI project, which introduces three innovative areas (AI Detector, Data Sanitization & Anonymization; AI Fairness), bringing them to TRL 8 means reaching a level of maturity that ensures solutions are fully validated and ready for use in concrete operational scenarios, offering reliability, performance, and practical applicability.”
Promoting safer AI: Datrix’s response with AIA Guard and STAR4AI
In a landscape where AI-related cybersecurity is still evolving, AIA Guard and STAR4AI represent a unique reference point in Europe, offering concrete solutions to overcome challenges related to data management and model transparency. With a focus on protecting Large Language Models (LLM), these technologies respond to increasingly pressing needs, such as preventing the unintentional sharing of sensitive data through commonly used tools like ChatGPT, ensuring not only the security of corporate data but also compliance with privacy regulations.
“Thanks to features like prompt sanitization and bias detection, AIA Guard and STAR4AI do not just protect data, but enhance the capabilities of data scientists, offering tools that improve the interpretability and transparency of models. This approach not only strengthens user trust in automated decisions but also allows companies to adopt artificial intelligence in a more conscious, fair, and strategic way” Shakirzyanova specifies:
In a context like the Italian one, where AI adoption is still limited, STAR4AI simplifies access to these technologies through intuitive and secure solutions. Practical benefits translate into tangible applications, such as personalizing models to improve customer care, using anonymized healthcare data to optimize diagnosis and treatments, and supporting strategic decisions in different sectors.
The result is an ecosystem that not only ensures security and data protection but also promotes innovation, breaking down barriers to adoption. A solution that could position Italian companies at the forefront, favoring their international competitiveness.
Embrace the AI Challenge: future challenges and opportunities for safer AI
Datrix’s mission is based on the idea that addressing the challenges of artificial intelligence is not just a necessity, but an opportunity to lead the change towards a safer, more transparent, and effective adoption of these technologies.
However, these same challenges pave the way for opportunities to expand market applications, lead in responsible AI adoption, and achieve continuous innovation thanks to the AI-humanity pairing. STAR4AI can position itself as a reference point for responsible AI adoption, thanks to technologies that promote transparency and trust, such as interpretability tools (XAI) and advanced bias detection. The ability to customize applications in specific fields – from healthcare to retail, to finance – allows transforming cutting-edge technologies into concrete and useful solutions, accelerating the spread of AI in key sectors.
It is not just about accepting AI challenges, but about embracing them, aiming to create new industrial standards for algorithmic justice and data protection, influencing the sector well beyond the boundaries of the project itself. This approach not only addresses current problems but defines a new paradigm in which security, transparency, and innovation coexist, favoring a digital transformation that puts ethics and sustainability at the center.
AIA Guard Platform
