New York RAISE Act sets frontier AI safety rules for big models

New York’s RAISE Act is the clearest example yet of a U.S. state trying to regulate the behavior of frontier AI developers, not just their downstream applications. By requiring firms like OpenAI, Google, Meta and DeepSeek to publish safety protocols, perform risk assessments and report serious incidents within 72 hours, Albany is importing playbooks from sectors like aviation and nuclear into AI development. ([techcrunch.com](https://techcrunch.com/2025/12/20/new-york-governor-kathy-hochul-signs-raise-act-to-regulate-ai-safety/?utm_source=openai))

This matters for the AGI race because it turns vague safety promises into enforceable obligations with million‑dollar penalties. For large labs, the cost of compliance will be small relative to training runs, but the precedent is huge: California and New York together now define a de facto baseline for U.S. frontier AI operations while federal efforts stall. ([axios.com](https://www.axios.com/2025/12/19/new-york-ai-safety-bill-hochul)) That baseline includes transparency about catastrophic‑risk scenarios and a regulator empowered to ask tough questions. Smaller labs may benefit if the law reins in the most aggressive scaling strategies of the biggest players, but they could also find themselves swept into similar rules if thresholds are lowered over time.

Globally, other jurisdictions now have a live template for how to regulate advanced models without banning them outright. That will shape where labs build data centers, how they structure safety teams and how quickly they push toward more agentic systems.