Comment by Cloud Security Alliance AI Safety Initiative

California’s SB 53 provides the most concrete existing disclosure mandate in the United States. For models trained above 10^26 floating-point operations, it requires safety incident reporting to California authorities within fifteen days, transparenc...

Constructive engagement should focus on three concrete needs: a clarification that AI-specific attack vectors including data poisoning, model manipulation, adversarial ML inputs, and supply chain compromise of model artifacts constitute qualifying cy...

Effective human oversight for agentic systems requires interrupt conditions—predefined thresholds at which agent execution is paused and human review is required—combined with real-time monitoring infrastructure capable of detecting anomalous action ...