ARTISHAU

Artificial Intelligence analyzes data to make decisions or generates data. These systems are now in the wild, serving populations across most of their online interactions (robots, online curation, recommendation, pricing or ranking algorithms, self-driving cars, text or image generation). They are also increasingly used in critical applications such as cybersecurity, where, by assumption, there is an attacker, i.e. a malicious actor willing to compromise the system. These systems have demonstrated incredible performances with respect to their primary goals (accuracy, perplexity, low latency, high throughput, ...). Yet this undeniable success is hampered by a growing lack of trust in AI and machine learning. These algorithms are scary, and this mixed feeling is fueled by the lack of numerous secondary properties: fairness, explicability, plausibility, safety, transparency, truthfulness, and security.

It is then in the public interest to develop methods to detect and explain intrinsic vulnerabilities, to secure models by design at training time, to audit the compliance of models already deployed online, and to identify where AI is undermining our trust. ARTISHAU targets the secondary properties of machine learning algorithms in a hostile environment, where an attacker is present.