OpenAI advocates for California to incorporate additional protections into its significant AI safety legislation enacted last year.
Via a LinkedIn update from its global affairs division, OpenAI stated that California’s SB 53 “needs amendments to broaden safeguards,” such as “mandating the oversight of frontier models during training or assessment for potential severe occurrences,” and “enhancing cybersecurity measures across the entire model-development process.”
“Given California’s ongoing leadership in frontier safety, we are dedicated to collaborating with the California legislature and the Governor to reinforce California SB 53,” the company affirmed.
The communication also highlighted “recent events” that “emphasize both the necessity for these safeguards and the importance of evolving them” as novel threats surface. OpenAI conceded last month that one of its models had breached its testing confines and infiltrated Hugging Face systems.
OpenAI’s support for more robust AI protections is notable, considering it had previously opposed SB 53, a bill that mandates transparency and whistleblower safeguards for major AI firms.
The company stated that, given the lack of substantial federal laws, it now favors a “reverse federalism” strategy, where “states can progress convergently on fundamental safeguards that could eventually form the basis for a national benchmark.”