Post by Nihan Yıldırım

AI Safety is the number one priority for all of us. AI is not limited to LLMs, but GenAI and Agentic AI play a major role in transformation today. However, many organisations face a common challenge: ➡️ How can regulatory requirements such as the EU AI Act be translated into practical, day-to-day AI governance and safety management? We are working on developing an AI Safety Assistant as a Decision-Support Framework for EU AI Act Compliance and Safety Evaluation with my students Öykü Şeker and Rana Talin Yavuz as a part of their Graduation project in ITU Management Engineering Programme. We have presented our initial findings on the development of an AI Safety Assistant system, including requirements and system proposal in #IAMOT2026 conference (and received high interest, honoured with a BPA). AI Safety Assistant framework is needed to bridge the gap between AI regulations and operational implementation. 🔍 Through a systematic literature review of 200 publications, industry benchmarking, and interviews with 18 stakeholders (AI users, practitioners, researchers, and policy experts), we identified five critical gaps in current AI safety practices: • Ambiguity in AI risk classification • Fragmented safety evaluation approaches • Difficulties in operationalising regulatory compliance • Challenges in selecting appropriate AI safety tools • Lack of traceability and explainability throughout the AI lifecycle 💡 Our proposed AI Safety Assistant integrates: ✅ AI risk assessment aligned with the EU AI Act ✅ Multi-dimensional safety evaluation (fairness, transparency, robustness, privacy, and security) ✅ Compliance mapping to regulatory obligations ✅ TOPSIS-based decision support for selecting AI safety tools ✅ Automated documentation and audit trail generation One particularly important finding was that users do not simply want compliance checklists. They need: • Context-aware guidance • Real-time feedback and warnings • Clear explanations of regulatory obligations • Support in verifying AI-generated outputs • Transparent decision-making processes Our research demonstrates that the main challenge in AI governance is not the absence of regulations, but rather the lack of mechanisms that translate regulatory requirements into actionable workflows. As organisations increasingly deploy AI systems, there is a growing need for practical tools that help ensure AI systems are not only innovative but also safe, trustworthy, and compliant.