MONACO DIGITAL: "Sovereign AI will be available in 2025"

Risk analysis, KYC, cybersecurity...Françoise Milatos, Deputy CEO of Monaco Digital, explains the challenges of AI for banks - and how Monaco Cloud's sovereign AI "integrating more efficient, latest-generation components" will be a game-changer.

Do you have a lot of requests from Monégasque banks to incorporate AI into their processes?
In many sectors, we are seeing a strong enthusiasm for artificial intelligence.  We can clearly see this at trade fairs: 80% of people come to talk about AI! This is also the case in the financial sector.

What are their main requests?
Recurring requests relate to Know Your Customer (KYC).  The objective is to better understand the customer and, for example, better measure the customer risk linked to the characteristics of financial products, thus refining offer and price proposed; financial institutions can also better anticipate and target client needs, even before they express them. 
The other major theme that comes up a lot is risk analysis (not only customer risk but also economic, political and even geopolitical risks - and of course also risks of fraud and cybersecurity attacks) and response strategies in relation to this set of risks.
For all these needs, AI-based systems use machine learning models to analyse thousands of documents and data, to better understand customers (for KYC) or to detect anomalies or signs of falsification and anticipate risks and attacks.  The results obtained with AI are impressive in speed and precision.

Nevertheless, we must be careful of biases to avoid a male smoker over 50 being mathematically excluded from credit, for example
Absolutely.  Governance and management are essential so that the models do not deviate from their main objective! We start by choosing the right models, specially Large Language Models (LLMs), the most suitable for the subject and context in the prototype phase.  There is then a phase of learning and refining the model to detect biases and reduce them by targeting the result.  This is part of the methodology.  Gradually, the model feeds on what we give it and the corrections we make to it.  The adaptation/refining phase is essential.

Is AI an absolute weapon in the detection of fraud and cyber attacks?
We now know that cybercriminals use a lot of AI in their attacks.  To counter this, auditors who carry out pen tests (penetration tests to find vulnerabilities) study all attack methods and include them in simulations.  This makes it possible to detect specific flaws that are detected by AI and to counter them with AI.  Red teams (ethical hackers who are experts in the attack methods of malicious hackers) are therefore increasingly specialising in AI, just as cybercriminals do, of course, in order to be able to uncover and subsequently counter these attacks.

Banks can use AI to set up trading strategies.  Does market analysis feature in your clients' requests?
We set up models that allow us to analyse data with predefined criteria.   Afterwards, the client feeds in the data to target their objectives.  We mainly have to deal with requests related to the targeting of products offered by financial institutions.  For example, a financial institution asked us to analyse environmental data to better refine the prices of these products and be more competitive.

Are you asked for black box audits? For example, to see what information about a company is visible on the Dark Web, or to understand how attackers use AI in their attacks?
Yes.  Pen tester auditors organise attacks on systems to detect flaws and then correct them.  Audits allow us to test the security measures in place.  This is a very important lesson for all companies, both small and large.  Banks are often better protected than companies in other sectors but there is no protection that is infallible!
In the Security Operations Centre (SOC), we accumulate all the data that comes from clients' information systems to isolate what we call weak signals, the signals that could be characteristic of an attack, an incident or bad practice.  In the analysis of this data, in the filtering, in the detection of false positives, AI can provide both greater acceleration and relevance; and also in rapid remediation techniques, we can automate a certain number of basic responses, when incidents are detected.

Artificial intelligence can only offer its full potential if companies ensure efficient data collection
We need relevant data and we need data that is well governed with all the associated regulations.  Monaco has now adopted the European law on the protection of personal data.  The European AI Act has not yet been adopted in the Principality.  It would be necessary to define a regulatory framework but also to adapt it to Monaco: we have the possibility of making a less rigid and better targeted regulation than the European AI Act, because we have better control over our environment. 

AI requires a lot of computing power, provided by Graphics Processing Units (GPUs).  Monaco Cloud has reserved its GPUs for the Monégasque government until now.  When will you offer them to private players?
The Monaco government has been using GPUs on Monaco Cloud for two years for certain projects developed by Extended Monaco.  We are currently working on setting up a sovereign AI environment, integrating more efficient latest-generation components.  This environment will provide models and tools that customers can use for their own needs.  The great advantage of a sovereign AI is that the data environment will be both secure and located in Monaco.

Is a Monégasque generative AI likely to be created?
It is not a question of creating the models and tools but of using those of our partners and customising them for Monaco.  We work a lot with Microsoft and IBM on this subject.  IBM has the particularity of being able to implement AI models that are localised.  It is imperative to control the environment that we are going to use so that it is protected and reliable, which is not the case for example with chatGPT.  By using an environment that is localised, we completely control the location of the processing and data.  We are in the phase of defining the architecture.  The sovereign AI will be available in 2025.

Will the sovereign AI be eco-friendly?
What we can say is that having a sovereign AI will also make it possible to better control consumption and to put good practices in place to reduce the carbon footprint.