Talk at AI:Advanced Workshop
The AI:Advanced Workshop, held on 5 November 2026 at TU Wien, brought together 30 participants and eight experts to explore the practical applications, benefits, and challenges of artificial intelligence. Through expert presentations and discussions, the workshop highlighted the importance of critical thinking, informed decision-making, and human responsibility in the effective use of AI.
Daniel Arp from TU Wien’s Security & AI Lab (SAIL) delivered a talk titled “Artificial Intelligence in Cybersecurity” (Künstliche Intelligenz in der Cybersicherheit). He explored how AI is reshaping cybersecurity, from emerging threats and AI-assisted cyberattacks to the challenges of using AI systems safely and responsibly.
The first part of the talk focused on how generative AI is changing the nature of cyber threats. Drawing on real-world incidents, including the Arup deepfake fraud case, Daniel demonstrated how AI-generated text, audio, and video can make phishing and social engineering attacks more convincing. He discussed the limitations of current deepfake detection methods and introduced approaches such as digital watermarking and content provenance verification. Daniel also examined the growing role of AI in both cyberattacks and defence. AI-powered agents can automate vulnerability discovery and facilitate sophisticated attacks, but they can also help security professionals identify vulnerabilities, develop patches, and improve software security. He emphasised the importance of timely updates, risk-based vulnerability management, and effective protective measures.
The second part of the talk addressed security and privacy risks associated with AI systems, particularly large language models and autonomous AI agents. Daniel explained how prompt injection attacks can manipulate AI behaviour, potentially exposing sensitive information or compromising system integrity. He outlined practical safeguards, including input and output validation, limiting AI system permissions, and maintaining human oversight.
Finally, Daniel compared different AI deployment options, including commercial cloud services, European AI providers, and locally hosted models, highlighting the trade-offs between performance, data protection, security, and organisational control.
photo: A. Rastegar/TU Wien Academy