Specificities of AI/ML solutions (in view of their software security) AI/ML systems are not developed and conceived by the application developer as (any) other data processing software is. The divergence relies on the core code (learning algorithm) genesis, not developed by the application developers but conversely by machine learning framework teams. Algorithm and framework are common bricks used for several very different applications by the AI/ML (vertical) solution designers. Some depicts this situatiŽŶ ĂƐ ͞ / ŝƐ ĂƵƚŽŵĂƚŝĐ Ɖ factual statement as no one needs to program it. A common pre-existing software is customized and tuned automatically at the training stage resulting in the production of the model parameters. The added value or intellectual property of the AI/ML vertical application designer is in the appropriate selection of the (pre-existing) algorithm and its training with the well- or best-fitted training data. A stock phrase is that AI/ML quality depends on its training data. As a consequence, with AI/ML, the ŝŶƚĞůůĞĐƚƵĂů ƉƌŽƉĞƌƚLJ ƌĞƐŝĚĞƐ ƉƌĞĐŝƐĞůLJ ŽŶ ƚŚĞ not an algorithm. This entails one critical Intellectual Property Rights (IPR) issue. The higher importance of data versus software for correct functioning of AI/ML and on the intellectual property value only fades partially the relevance of software security. In the context of an open-source AI/ML algorithm code, software confidentiality is still surprisingly important as a properly targeted attack on an AI/ML solution is far easier to spawn with the accurate knowledge of algorithm. In practice, there are many of these algorithms. Keeping the algorithm confidential is not aimed at securing the intellectual property but to conceal and hide it, thus creating opacity on its actual type to the attacker, drifting the attack from a defined strategy based on certain assumptions into guesses. On the other side, considering the model parameter security, it is at least equally important to elaborate how these data can be protected leveraging software security in appropriate schemes. In practice, securing the software which itself secures and validates the data is a valuable security boost. When software security is discussed, a common statement is that open-source code is allegedly safer as being (security-wise) improved by a massive class of users. The rationale at stake is a lower prevalence of pending vulnerabilities to exploit than in proprietary code of lesser use. This argument is a dual sword argument as hackers wil...
