Cloud Security Alliance AI Safety Initiative

Third, embedding addresses a real information problem. Frontier AI risk assessment depends on non-public facts such as training details, internal monitoring data, and incident records. An evaluator that can see only what the developer chooses to send...

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...

The rapid commercialization of artificial intelligence has produced a market structure that bears an uncomfortable resemblance to the conditions regulators spent decades trying to prevent in telecommunications and financial services: a small number o...

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 ...