Can the US and China Stop AI From Killing Us All? The Race to Control Artificial Intelligence

 Can the US and China Stop AI From Killing Us All? The Race to Control Artificial Intelligence

Can the US and China Actually Team Up to Keep AI From Going Off the Rails?

It sounds like the plot of a movie: the world's two biggest AI superpowers, locked in fierce competition with each other, somehow also need to cooperate to make sure the technology they're racing to build doesn't spiral out of anyone's control. But that's genuinely the situation the US and China find themselves in right now, and there's actually some real movement on it.

Why This Conversation Is Happening Now

For about two years, official AI safety talks between Washington and Beijing had basically gone quiet. But both sides recently agreed to pick things back up, with a dialogue reportedly planned for around mid-September, possibly timed near a summit between President Trump and President Xi Jinping. And there's a pretty specific reason the urgency has picked up: AI models have gotten a lot more capable, a lot faster than most people expected.

One moment that seems to have sharpened everyone's focus was Anthropic's announcement back in April that its newer model had discovered previously unknown software vulnerabilities on its own — capable enough that the company held off releasing it publicly and instead built a collaborative security framework with partner organizations to help harden systems against the kinds of attacks that model could enable. That's the kind of development that tends to make policymakers on both sides sit up and pay attention, because a sufficiently capable AI system doesn't really respect borders. A cyberattack it helps carry out, or a dangerous technical capability it makes possible, doesn't stay neatly contained within the country where the model was built.

What They're Actually Trying to Agree On

To be clear, nobody's expecting the US and China to sign some grand arms-control-style treaty on AI. The scope being discussed is much narrower and more practical: things like monitoring AI-directed cyberattacks, encouraging AI labs on both sides to share information about emerging threats, and finding enough shared visibility into what each side is genuinely worried about so that mutual suspicion doesn't turn into an excuse for reckless, unchecked development.

Interestingly, China isn't just sitting back on this either. Its own regulators have already built loss-of-control scenarios into official AI safety frameworks, and Chinese research organizations have published their own catastrophic-risk frameworks, complete with tiered warning systems for when a model's capabilities start crossing into genuinely dangerous territory. So there does seem to be some overlapping concern here, even amid all the competitive tension.

The Honest Complication

Here's the part that makes this tricky: trust is in short supply. Washington has accused Chinese firms of appropriating US technology, while Beijing pushes back against what it sees as American attempts to monopolize the most advanced AI systems and write all the global rules in its own favor. Layer that on top of a broader economic and strategic rivalry between the two countries, and you get talks that are trying to build cooperation on a genuinely difficult foundation.

There's also a real debate about whether meaningful cooperation with China on frontier AI risk is even possible in the first place. Some analysts are skeptical it can work at all, while others argue that even a "narrow opening" — some shared visibility, some basic lines of communication — would be a meaningful improvement over total silence, especially if the two countries can keep these safety conversations insulated from the rest of their economic sparring.

Why It Might Matter More Than It Sounds

The case for taking this seriously comes down to a pretty simple idea: a dangerous AI incident doesn't check passports. An autonomous system exploiting a vulnerability, or an AI-assisted cyberattack, can cross networks and borders in seconds, long before any government has time to react. That's part of why some experts frame these upcoming talks as coming at, in one analyst's words, "the most critical juncture" — arguing that this window for building even minimal cooperation might not stay open indefinitely.

So, Can They Stop AI From Killing Us All?

Honestly? Nobody can promise that, and reasonable people disagree pretty sharply on how large that risk even is. But what does seem clear is that both countries have enough of their own reasons to want some guardrails in place — even if they can't agree on much else. Whether these talks turn into anything durable is still an open question. But the fact that the world's two AI superpowers are even sitting down to talk about it, after two years of near silence, feels like a small, cautiously hopeful sign in a conversation that badly needed one.