The concept of robustness is one of the fundamental pillars of trustworthy AI. AI systems must be developed to ensure an appropriate level of accuracy, robustness, and cybersecurity. We are committed to developing AI systems that are both safe and secure while reducing the risk of failure or misuse.

To ensure that its AI systems are robust and safe, Lefebvre has adopted a rigorous “Security by Design” approach.

To ensure the proper functioning of its AI systems and the protection of the information they hold and process, Lefebvre Group is committed to implementing the following measures:

• Data security;
• AI model security;
• Security of the infrastructure supporting the AI systems;
• Security of codes and algorithms;
• Security of learning processes;
• Security of user interfaces (UI) and Application Programming Interfaces (APIs);
• Security of deployment and operations.

More broadly, Lefebvre Group strives to adopt the preventive measures required against malicious manipulations and potential security vulnerabilities that may affect its AI systems.

We address the security of AI systems holistically by protecting data, machine learning models, infrastructures, software, and ensuring regulatory and ethical compliance throughout all stages of design, development, deployment, and maintenance of these systems.

The energy footprint of digital technology, especially artificial intelligence technologies, is a major concern for Lefebvre Group, since AIs are now at the heart of our core value.

We intend to apply this vigilance when designing and implementing AIs on one hand, and in empowering our users, collaborators, and clients on best practices on the other hand. This is what we mean by reasonable use of AI.

Design and engineering: our priority is to provide efficient programming, with a low-resource footprint when developing AI systems. We require the same high standards to our suppliers’ solutions or services.

Use: we pay close attention to the behavioural changes driven by generative AIs, and we wish to support our clients in adopting a measured use of these technologies.

During the design phase, we do not use personal data, which means we do not use any personal data for training purposes.

When deploying AI systems, we stand for Privacy by design: data protection principles must be integrated from the very beginning of any AI project. This ensures that personal data is processed in a compliant and secure manner from the start. System settings must by default ensure the highest level of confidentiality.

However, when using AI systems, personal data may be used. In this case, the connection between the General Data Protection Regulation (GDPR) and AI is clear. We emphasise the lawfulness, fairness, and proportionality of processing. To this end, we commit to respecting the following rules:

• Use of compliant data: The data used in the context of these AIs must have been collected in compliance with data protection requirements and must not be used for illegitimate purposes.
• Data minimisation: Personal data collected for the use of these AIs must be limited to what is necessary to achieve the intended goal. Thus, an evaluation of the amount of data must be done regularly to reduce the number of data used during the learning phase and that used during the use of the systems.
• Enhanced transparency: we strive to develop procedures to make AI systems transparent and understandable (see next principle).
• Data security: The confidentiality and integrity of information are paramount for our clients as well as for us. We thus implement adequate security measures to protect personal data against any unauthorised access, disclosure, or alteration. In the context of an ethical and respectful use of our AI systems, we commit to ensuring the protection of users’ personal data. Furthermore, we commit to ensuring the protection of confidential information and trade secrets.

The concepts of transparency and explainability when we communicate with our users are key to ensuring trust in our AI systems.

This involves informing users about the personal data or business data collected and used, as well as the decision-making modalities of AI systems.

During the design or deployment phase

Transparency is essential to ensure that decisions made by AI systems are understandable. This includes the ability to clearly explain the mechanisms and decision-making processes of an AI system.

“AI transparency” thus involves providing users with all the information related to the design, operation, and decision-making processes of an AI system which, when made available to users, allows them to observe its operation, enabling them to understand it, as well as the scope of the data that were used.

As a matter of fact, our AI systems are made transparent about the algorithms and modeling principles that are used, the process operated by the machine, and the products resulting from this process.

This transparency in AI systems also requires identifying the data used, whether personal or not: training data in the case of machine learning, repositories, or ontologies for symbolic AI. We are committed to providing clear explanations on how the system uses data. Users must be fully informed of the data collected, how it is processed, and their rights.

Lefebvre Group will apply the Code of Conduct recommended by the European Commission’s Artificial Intelligence Bureau on these points.

During use

We commit to explicitly notifying the use of an AI system whenever this is the case, whether the user is interacting directly with an AI system or when using the product obtained through the execution of an AI system.

To do this, Lefebvre Group commits to using appropriate interface elements or visualisation processes.

The design of AI systems, whether based on symbolic approaches or machine learning, must adhere to a quality and business requirement framework established through collaboration with relevant professionals. This process involves iterative dialogue and evaluation by these professionals.

This ensures optimal reliability of AI systems in their understanding of business reality, minimising the risk of gaps, biases, or even hallucination in the case of generative AI.

Furthermore, Lefebvre Group commits to designing and using AI systems that provide the means for understanding the reasoning of AI, and which thereby encourage users to a non-ambiguous understanding of the process operated, and a porper awareness of all the steps in the process.

The AI systems we design or use should also allow the users to understand and control the processing of their data and to be aware at each step of the implications related to this use.

We make every effort to protect the value of the content modeled and trained by our AI solutions, ensuring the reliability and quality of the resulting products.

We actively participate in ongoing discussions about the impacts of AI on copyright rules and contribute to defining best practices for the overall knowledge economy, including the press, media, and publishing industries.