And security training awareness is something we take very seriously.

So don't think it's gone unnoticed that you led the Knicks to an NBA championship, first time since 1970.

And yes, the 5-0 49ers as we record this.

Yeah, I hope we didn't jinx it.

I know, knock on wood.

I wish I could talk about the 49ers all day, but to get a little more broad here, there seems to be kind of a war in AI capabilities, right? So AI is making social engineering easier to pull off, but it also is significantly improving defensives.

Can you talk about kind of the battle lines there and how you as an organization try to use AI and technology to keep up with the AI-enabled threat landscape?

And it will continue to be the long tail of problems even as you harden your infrastructure.

But from where you sit, what should we be paying attention to right now, and what is super important?

Well, we're coming out of the summer of doom, right? I think this is the summer where the threat of AI and rogue AI agents has come into the public consciousness in a way that it just hasn't.

You know, I have relatives in my life who I've never talked about AI before who are saying, "Is this thing really gonna kill us?" And my honest answer is, "Maybe." [chuckles] So what's really interested me is just as a lawyer, how we can use our levers, both at the federal and state level, to reform liability law.

'Cause you have these big frontier models, the OpenAIs, the Anthropics, out there saying, "This actually might kill us all." And it's, you know, it's the engineers, sure, but it's also the CEOs who have put out these pretty frightening statements saying, "If we don't properly align ourselves with human values, like this thing could go off the rail pretty quickly." So I think there's a lot of room and a lot of conversation about changing liability law.

It's gonna be difficult.

We generally ascribe liability either through negligence or strict products liability.

In negligence, you need to actually go in and prove that somebody was negligent, and I pride myself on knowing a little bit about the tech.

If, if you, if you don't do it, you're not gonna get the reward.

Yeah, for sure

...

I think Hugging Face was the warning shot to people in the know, people in the community who understand what Hugging Face was, who are technologically adept to realize, like, okay, they escaped the sandbox, uh, and they, uh, exceeded their authority, and they were communicating with one another, and we're all gonna die.

And I think that was understandable to people who are in the field, but this was...

this is where it broke containment, to use the same term, in terms of the public consciousness.

And I have people in my life who I've never really talked to about this stuff being like, "What is, what is going on?" And then I think Amadei, um, and Altman this weekend were reacting to that reaction, and there was just this understood need to, to do something.

I mean, there's a lot of responsibility on these frontier models, and I think the pressure was growing for them to, uh, to put a pause on this.

Right.

Oh, 'cause you're putting it in voluntarily.

Yeah, you're kinda forfeiting your expectation of privacy.

Uh, you know or should know that, uh, the model that you're using is gonna retain that information.

And so unlike, say, the contents of an email, which you do have a reasonable expectation of privacy there because it's supposed to be a private communication between you and another person, so far at least, that's not how courts are seeing chatbots.

So it creates this really interesting dynamic where people are telling pretty secretive things to their chatbots and not realizing that that's gonna be used against them, not just in criminal cases, but in civil cases.

So this Washington Post exposé lists a few anecdotes.

Some of these are, are kind of funny actually.

Mm-hmm.

Hmm.

So there's kind of the mens rea or criminal, uh, intent element here that the reason you are destroying that property is for the purpose of preventing government access to something, and here that would be the device.

What makes this case unique is that all of the experts interviewed for this article were unaware of any previous case in which prosecutors had targeted a person because they used Graphene or similar, uh, privacy-focused operating systems to wipe their phones in a scenario like this.

And this could have significant downstream effects.

If the use of these, uh, Graphene-type tools is enough to cause law enforcement suspicion, then people who wanna keep their communications private for whatever reason, uh, they mentioned journalists or activists or lawyers, or people who are a part of disfavored religious or, or, uh, other types of groups, people who are using privacy tools and security tools like Graphene for their own purposes might be discouraged from doing so, and that would, uh, have a v- a very deleterious, uh, impact.

That goes against best practices.

It goes against the recommendations of security experts who will tell people to employ these tools for their own protection.

Mm-hmm.

Mm-hmm.

And now that is extending into the AI realm, where politicians are trying to manipulate chatbots into saying positive things about them.

So the hook for this story is about a Democratic candidate for the Missouri State Legislature named Dustin Lloyd.

Uh, and in doing research, he figured out that AI chatbots didn't really know anything about him.

Uh, there was some basic public information about who he is and what he's running for, but the chatbot knew nothing about his policy priorities, so he decided to expand his own website.

He put a question and answer section in there about his background and his goals, uh, and, uh, his, his policy prescriptions.

And almost instantaneously, he found that the chatbot responses were becoming more detailed.

Hmm.

Mm.

Part of a new strategy on the part of the European Union through the European Commission to achieve technological sovereignty.

It is designed to reduce Europe's dependence on foreign, and particularly US cloud providers, uh, and strengthen European technology autonomy within their own jurisdictions.

And I think you have to think broader than just technology here.

We've seen this in other contexts where leaders of other Western democracies have talked very openly about the inability to rely on the United States as a provider of X, Y, and Z.

One of the speeches that I think was the major catalyst for this type of thinking was one that, uh, Canadian Prime Minister Mark Carney gave in front of, uh, the Davos conference in Switzerland, where he talked about how, given what's happened with the Trump presidency and the fact that he was elected two separate times, countries need to prepare for the fact that they can't rely on the United States to be the provider of military technology or collective defense, that countries are gonna have to assume sovereignty over these domains where they've been overly reliant on the United States, and one of those spheres is technology.

And I think that's really the context that we need to understand why they're doing this.

Uh, do you think this administration, uh, has an, an open ear to those sorts of complaints?

Hmm.

So the Supreme Court weighed in on one of those questions in a, uh, groundbreaking decision yesterday.

They said that this is a Fourth Amendment search, and therefore a warrant based on probable cause and a particularized warrant is required in these types of cases.

It was a decision authored by Justice Kagan.

She was joined with, uh, the two other liberals on the court, Justice Jackson and Justice Sotomayor, as well as the chief justice and Justice Kavanaugh.

The crux of the opinion is that a person has a reasonable expectation of privacy in, uh, their location data that they share with Google.

Uh, every time we use a Google application for the first time, we get that little pop-up saying, uh, "Do you affirmatively agre- uh, agree to use location services? Um, it's really gonna improve your in-app experience," et cetera.

Uh, how does this affect, uh, everyday folks as well as the technology companies and, and law enforcement as well?

Mm-hmm.

Uh, the controversial aspect of it is, uh, the fact that there's incidental collection of US persons communications.

Those go into a database that can be searched without a warrant, uh, in many different contexts.

So there have been some significant fights on reauthorization.

There are members of both political parties who are adamantly opposed to Section 702 in the absence of a robust warrant requirement for querying of that database.

But when push comes to shove, Congress has always reauthorized Section 702 when it's been set to expire.

Right.

They authori- reauthorized it easily in 2012.

Can you bring us up to date here, Ben?

Sure.

So just to catch everybody up, Anthropic had a deal with the Pentagon.

They were gonna be involved in a bunch of different Department of War, I guess I should call it, uh, activities.

Uh, but they tried to put up some guardrails saying that they weren't gonna be fully compliant when it came to, uh, fully autonomous weapons systems and mass domestic surveillance.

Uh, and that disagreement came out into public view at the end of February, and then the Pentagon not only decided to terminate the contract with Anthropic, but the administration issued this executive order saying that because Anthropic now represents a risk to the supply chain, any federal agency and any federal contractor is prohibited from doing business with Anthropic.

And the particular contract with the Pentagon was replaced, uh, with a contract, uh, with OpenAI, so a, a different company.

Hmm.

[laughs]

Yeah.

If any, any sponsors wanna, uh, you know, help us out with that, we can give you firsthand knowledge.

But, um, so that, that's kinda one element of it.

But then this question of, like, what exactly did OpenAI agree to, and why potentially should we be suspicious? Uh, f- so from what OpenAI said is they, uh, guaranteed safeguards to protect against domestic surveillance and to protect against, uh, the use of autonomous weapons systems, which if that sounds very familiar, it is.

If you read between the lines, the difference is that OpenAI is accepting a legal framework that references existing laws and Pentagon policies.

They're not looking for these contractual assurances that Anthropic was looking for, which says, like, "You need to come to us if you are intending to do these things, and we are gonna be very protective of our models and institute our own human controls." OpenAI is just like, "We promise we won't break any laws.

We also would never.

Okay.

Uh, so you do not have to have probable cause for an administrative, uh, subpoena in any context.

You just have to have reasonable suspicion, which is a lower standard.

Uh, and administrative, uh, subpoenas are used for all different types of things.

We see them frequently in the national security context.

There's a, a specific administrative subpoena tool called National Security Letters.

Those are even more severe than the type of subpoena we're talking about here because those orders usually go to individuals, uh, who work for big tech companies.

Can you just give me, give us the 101 on the difference?

Hmm.

So the allegation is that, um, in the way these products are designed, they produce these sort of addictive features, so infinite scroll, autoplay, persistent notifications, et cetera.

They've pointed to some literature saying that these tools in social media have caused or worsened depression, eating disorders, self-harm, uh, in the most extreme cases, suicide.

And they are going to go through the discovery process to try and prove that these companies knew or should have known about harms to minors but failed to act.

This is gonna be a very difficult lawsuit for the plaintiff.

It's gonna be very difficult, uh, for them to, uh, prevail on the merits here.

This is being considered under a products liability framework.

No, not, he's not at his best in those situations it, it seems to me.