CAIP Celebrates the International Network of AI Safety Institutes

Last week, the United States hosted the inaugural meeting of a global network of Artificial Intelligence Safety Institutes, with delegates from Australia, Canada, the UK, the US, the EU, Japan, Korea, Singapore, and Kenya.

The delegates pledged $11 million in funding to research safeguards to protect against nonconsensual deepfakes and fraud, and they conducted a first-of-its-kind multilingual model evaluation, which has already yielded insights that will help deliver “more robust and reproducible AI safety testing across languages, cultures, and contexts.”

The Center for AI Policy (CAIP) celebrates this meeting and encourages the delegates to continue their important work. The challenges posed by AI cannot be contained within national borders – we have no alternative but to work together to solve those challenges.

As Dario Amodei, CEO of leading AI developer Anthropic pointed out at the meeting, one of these challenges is finding a way to make sure companies are actually living up to the letter and spirit of their voluntary safety commitments. “I think we absolutely have to make the testing mandatory,” he said, cautioning that without mandatory testing requirements, “there’s nothing to really verify or ensure the companies are really following” their responsible scaling plans. 

Although the testing requirements will need some flexibility because of how quickly AI technology is changing, that doesn’t mean developers should be left entirely to their own devices.

The dangers are simply too intense for us to place blind trust in America’s AI companies. CAIP is pleased to see this recognized by US Commerce Secretary Gina Raimondo, who opened the meeting by warning of “the possibility of human extinction.”

“Why would we choose to allow AI to replace us?” she asked. “Why would we choose to allow the deployment of AI that will cause widespread unemployment and societal disruption that goes along with it? … We shouldn’t. In fact, I would argue we have an obligation to keep our eyes at every step wide open to those risks and prevent them from happening. And let’s not let our ambition blind us and allow us to sleepwalk into our own undoing.”

Indeed, the stakes are nothing less than humanity’s undoing – but if the US leads the world in a cooperative effort to develop, require, and implement better AI safety evaluations, then we can avoid this fate. CAIP calls on Congress and the incoming Trump Administration to continue to support the US AI Safety Institute so that they can maintain their leadership over this vital global effort.