James Davenport

The EU AI Act places high importance on cybersecurity of AI systems and products, but there is comparatively little work done on this, and none that has reached the level of mature standards. Hence it is important to develop these standards, and ensure that they reflect both the cybersecurity point of view and the specific difficulties of AI, as in the ETSI list , and possibly wider.

Europe has already seen many cybersecurity attacks, whether by hostile nation states or by criminal gangs, even before AI becomes widely deployed. The impact of these has already led to at least one death, as well as much damage and distress. As AI becomes more widely deployed, these risks will only grow, and need effective standards-driven mitigations. The impact of my work will be coherence between the developing European standards in ISO-IEC JTC/1 SC27 and the current international draft standards in the area of cybersecurity. In addition, I will feed in research from the cybersecurity community as it affects AI-specific attack methods.