Frontier AI and the New AI Executive Order with Josh Geltzer

So first, the White House has issued a new executive order that arguably modifies the earlier EO on AI development.

We will add a hyperlink to that new EO in the notes to our cast.

But the order directs agencies to develop a voluntary framework for securing frontier AI models within sixty days.

Wow, that's fast.

Under the framework, AI developers will provide the federal government with access to leading-edge frontier models thirty days before they release the models to any other organizations.

The framework will also allow AI developers to work with the government to, this is a quote, "select trusted partners that will have early access to covered frontier models to promote secure innovation and strengthen the cybersecurity of critical infrastructure." This all sounds positive, right? Trump's order specifically bars EO language from being used to create mandatory requirements for AI developers to pre-clear their models with the government.

I imagine that was a nod to the private sector, but there may be also good reasons to do that.

Leaders at the National Security Agency, the Cybersecurity and Infrastructure Security Agency, the CISA, and the National Institutes of Standards and Technology, we'll call it NIST, among other agencies, are part of the group that is supposed to develop this voluntary framework, and these agencies are also tasked with creating a classified, quote, "benchmarking process" to determine when an AI model meets the threshold for voluntary review.

So among other things, this EO tells the Treasury Department to work with the NSA and the CISA and other agencies to form an AI cybersecurity clearinghouse.

The group was supposed to then work with the industry, work with operators of critical infrastructure, you know, many of those are actually private sector companies, to coordinate on new software vulnerabilities, and what I think might be interesting is, are they talking about the electrical grid? Are they talking about protecting nuclear power plants? If so, this seems like the right direction.

So we're gonna keep an eye on that, but most importantly, we're gonna talk to someone who knows a lot more about it than me in just a few minutes.

So first, the White House has issued a new executive order that arguably modifies the earlier EO on AI development.

We will add a hyperlink to that new EO in the notes to our cast.

But the order directs agencies to develop a voluntary framework for securing frontier AI models within sixty days.

Wow, that's fast.

Under the framework, AI developers will provide the federal government with access to leading-edge frontier models thirty days before they release the models to any other organizations.

The framework will also allow AI developers to work with the government to, this is a quote, "select trusted partners that will have early access to covered frontier models to promote secure innovation and strengthen the cybersecurity of critical infrastructure." This all sounds positive, right? Trump's order specifically bars EO language from being used to create mandatory requirements for AI developers to pre-clear their models with the government.

I imagine that was a nod to the private sector, but there may be also good reasons to do that.

Leaders at the National Security Agency, the Cybersecurity and Infrastructure Security Agency, the CISA, and the National Institutes of Standards and Technology, we'll call it NIST, among other agencies, are part of the group that is supposed to develop this voluntary framework, and these agencies are also tasked with creating a classified, quote, "benchmarking process" to determine when an AI model meets the threshold for voluntary review.

So among other things, this EO tells the Treasury Department to work with the NSA and the CISA and other agencies to form an AI cybersecurity clearinghouse.

The group was supposed to then work with the industry, work with operators of critical infrastructure, you know, many of those are actually private sector companies, to coordinate on new software vulnerabilities, and what I think might be interesting is, are they talking about the electrical grid? Are they talking about protecting nuclear power plants? If so, this seems like the right direction.

So we're gonna keep an eye on that, but most importantly, we're gonna talk to someone who knows a lot more about it than me in just a few minutes.