AI Safety is Stuck in Technical Terms -- A System Safety Response to the International AI Safety Report

Technical fixes alone are insufficient: The paper argues that current AI safety efforts are overly focused on technical solutions, but the International AI Safety Report and broader governance frameworks neglect the sociotechnical nature of AI risks—meaning safety requires addressing both technical and non-technical factors (like organizational practices, human decision-making, and social context) working together.

System safety offers a proven alternative approach: Drawing from decades of experience managing safety in software-based systems, system safety discipline integrates technical and non-technical considerations from the start rather than treating them as separate concerns, providing a more comprehensive framework than current AI safety discussions offer.

Current policy frameworks have blind spots: The paper identifies fundamental issues in how the European AI Act and international safety reports are structured, suggesting they may be inadequate because they're built on the same limited technical framing that dominates AI safety discourse, potentially leaving real-world risks unaddressed.

Meaningful safety discourse is being frustrated: The dominance of technical terminology and approaches in AI safety—even among 96 international experts—is actively preventing comprehensive discussion of safety challenges and limiting the effectiveness of policy efforts to address them holistically.

Interdisciplinary integration is essential, not optional: Rather than simply adding non-technical considerations to existing technical frameworks, the paper argues that system safety methods must be foundational to how we understand and govern AI safety from the beginning.