A new report from RISE Research Institutes of Sweden, supported by the CitCom.ai Testing and Experimentation Facility, provides an open analysis of security risks both targeting AI systems and enabled by them.
Researchers at RISE have published a survey examining the dual cybersecurity challenge posed by artificial intelligence: AI systems can be attacked and compromised, and they can also be weaponised to launch and scale cyberattacks. The report, co-funded by CitCom.ai as part of its mission to advance trustworthy AI across Europe, draws on 212 academic and industry sources published between 2021 and 2025.
The survey identifies adversarial attacks, data poisoning, model inversion, and data leakage as the primary threats facing AI systems — all of which exploit the data-driven, adaptive nature of AI in ways that conventional IT defences were not designed to handle. On the offensive side, the authors document how AI is already being used to generate personalised phishing emails at scale, produce deepfake media for fraud and disinformation, and create adaptive malware that evades signature-based detection.
The report also reviews the leading frameworks for AI risk governance — including the NIST AI Risk Management Framework, ISO/IEC 23894, MITRE ATLAS, and the OWASP Machine Learning Security Top 10 — and maps a range of threat actors, from nation-states pursuing strategic advantage to insiders and unintentional design flaws.
The CitCom.ai TEF, co-funded by the European Union, supports exactly this kind of applied research into AI trustworthiness and security. The survey is a resource for organisations assessing their AI risk posture, and for policymakers working to align technical safeguards with the requirements of the EU AI Act and related governance frameworks.
