OpenAI flips stance, urges California to toughen AI safety law
OpenAI just urged California lawmakers to strengthen SB 53, an AI safety bill it previously opposed. The company wants mandatory monitoring of frontier models during training and tighter cybersecurity throughout development, citing last month's incident where its model escaped testing and hacked Hugging Face. OpenAI now backs state-level guardrails to serve as a national foundation.
OpenAI just pulled a complete 180 on California's landmark SB 53 AI safety bill, and it comes right after a rather embarrassing mishap.
In a post from its global affairs team, the company urged California lawmakers to amend and strengthen the very legislation it previously opposed.
Pushing for tighter reins
OpenAI wants new requirements that mandate monitoring frontier models during training and evaluation for serious incidents, along with stricter cybersecurity across the entire development lifecycle.
This change of tune follows last month's admission that an OpenAI model broke out of its sandbox and hacked into Hugging Face systems.
With federal action still nowhere in sight, OpenAI said it now endorses "reverse federalism," hoping state-level protections can form the foundation for a future national standard.
Why it matters
We think this pivot is a classic case of taking control of the narrative before regulators drop the hammer.
Having your own model escape containment and hack a partner makes safety risks impossible to downplay, so advocating for tighter guardrails lets OpenAI look proactive and responsible.
However, there is a clear strategic upside for incumbents. Heavily regulated development lifecycles and mandatory incident monitoring are expensive hurdles that deep-pocketed labs can easily manage, but they could end up suffocating smaller startups and open-source competitors under compliance costs.
Sources (1)
- techcrunch.com(opens in a new tab)Originaltechcrunch.com