| This interview has been edited for brevity and clarity. Q: Your bill empowers the private sector to share information about model standards and evaluation with their Chinese counterparts. Can you describe the process you’re envisioning — and the legal carveouts it requires? A: So first, the rationale [is that] AI expertise overwhelmingly lies outside of government. This is not like negotiating over nuclear stockpiles. So we need to ensure that the experts are engaged with their counterparts in China in establishing the metrics and protocols for testing and evaluation — and, most importantly, the methods for verification to set the table for any future negotiation between the political leadership. And I have been told that there are some impediments to those conversations happening today — although they do happen, but not to the extent they can or should. And there are a series of laws and rules that create challenges: national security laws, which constrain how we can communicate to the Chinese about antitrust [since] we’re talking about private companies speaking to one another about safety, about safeguards; or even export controls. The legislation attempts to provide some certainty and confidence to those experts who are willing to engage in these conversations by having them register with the State Department, so the government knows exactly what conversations are happening and who’s having them. And there is essentially safe harbor from civil or criminal liability to be engaged in these conversations. We don’t trust the Chinese. We need to ensure we have clear methods of verification, and that’s not going to happen because somebody in the White House or Department of Treasury dreamed it up. It’s going to happen because technical experts understand exactly how to verify, and how reliable that verification will be. Q: Yes, I wanted to ask you directly about verification mechanisms. What would you say to someone skeptical that the Chinese will ever be truthful about the information that they provide? A: I’m obviously not the expert, so I don’t want to pretend to be. But, if you’re thinking about, for example, whether or not a model has exceeded a certain threshold of capabilities, one example of verification that some experts have discussed … has been the ability to be able to monitor how compute is used and whether it’s being used for model training versus model inference. What we also know is that China is already engaged in regulation in lots of ways. Look, I understand the Chinese may lie about lots of things, but what we do know is that they are already regulating models. They’re regulating them around deepfakes and content [and] around child protection. They’re increasingly concerned about loss of control, so we know they’re taking on issues of agentic misalignment, and they’ve been regulating for a while. Q: So, the bill says verification has to take place, but it’s left for the experts to figure out the details for later? A: Correct. So, the point is to have what they call track two negotiations. In international negotiations, the track two is generally technical experts who are at the table. Those are typically government officials. In this case, they would not necessarily be, because that’s not the expertise, and they would be discussing with their Chinese counterparts exactly how verification would work. Q: Could you speak generally to why you think it’s necessary to put out a bill like this right now? I see the proposal says that your office has no “delusions” about the administration’s willingness to come to the table on AI safety regulation. A: Despite what Trump has said about AI safety concerns being a hoax, what we see very clearly from many other members of his administration is [that they] are quite concerned about safety. In the last 24 hours, we’ve had both the secretary of the Treasury and secretary of state talking about having a red phone, so there’s no question there is concern among some in the administration. There’s clearly concern within the industry. We hear that from the chief scientists and the CEOs themselves. We’re hearing lots of concern from corporate leaders. We just had an experiment where Southern California researchers demonstrated how an AI agent would break into WeChat, and 1.4 billion account users would then be vulnerable. Again, that’s an experiment. Q: You do have a Republican co-sponsor on this bill despite the anti-regulation rhetoric coming from the administration. What are you sensing from your private conversations or deliberations with your Republican colleagues about where they are on this issue? A: I have been hearing positive signs in my conversations with several Republican colleagues that they get it, and, most importantly, that their constituents get it — and they need to affirmatively respond to the reasonable fears that they are seeing in their own communities. There’s understandable reluctance from several Republicans to engage directly with China. I’d like to believe if we can get past Election Day, the more rational instincts will encourage them to engage. This newsletter is published by WP Intelligence, The Washington Post’s subscription service for professionals that provides business, policy and thought leaders with actionable insights. WP Intelligence operates independently from The Washington Post newsroom. Learn more about WP Intelligence. | 
