Insight

AI Armageddon Looks Different From China

By Catherine Thorbecke Bloomberg View

Published Sept. 15, 2026

Embed from Getty Images

Article Tools

Contact The Editor Articles By This Author

Rather than rogue artificial intelligence agents supposedly killing us all, Beijing has a more immediate nightmare: That the technology could help humans tear society apart first.

After Silicon Valley spent the past week ramping up fears of AI destroying humanity, China rebuffed what it sees as "fearmongering." Minister of State Security Chen Yixin separately penned an essay on Sunday offering a glimpse into some of the nation's biggest fears. It doesn't mention human extinction. The top concern is how the technology that could ruin social stability via deepfakes, influence operations, and cognitive warfare at scale. Second was cyber threats that could make it much easier to harvest sensitive and personal information.

A recent episode suggests those two risks could go hand in hand. Earlier this month, a small team of cybersecurity researchers in California published findings showing how they were able to simulate a major breach of WeChat, the super-app that essentially serves as the digital critical infrastructure for daily life in China. Calif, the Palo Alto-based company that published the so-called "WeWorm" threat, confirmed that WeChat's parent Tencent Holdings Ltd. has mitigated the risk.

But their demo is nonetheless unnerving enough to keep Beijing officials up at night: The team detailed how they could use AI tools to create a bug that was able to take full control of a WeChat account in seconds. All an ordinary user sees is an incoming call from someone on their contact list.

Still, there's reason for hope. The disclosure with Tencent went smoothly and there were no technical hiccups, Calif's Bruce Dang told me over email, adding that the engineers were responsive and the issues were "quickly fixed." A disaster that could destabilize a platform used by more than a billion people to run businesses, pay for goods and communicate was thwarted because of quick cross-border collaboration.

In a blogpost, Calif Chief Executive Officer Thai Duong described the experiment as a "call for the United States, China, and other governments to work together and collaborate with private industry on developing and deploying AI" safely. Most notably, he added that blaming AI and curtailing its further development is "the wrong lesson." The vulnerabilities are already out there, what's changed is that AI can "find and fix them." In other words, the WeChat episode isn't about slowing down the race, but making transparency a viable path forward that is politically acceptable in Washington and Beijing.

Despite the rising apocalypse warnings from Silicon Valley, President Xi Jinping emphasized Beijing's pitch for an "open ecosystem for AI" during remarks at the BRICS summit in New Delhi on Sunday. Xi called for a "consensus-based" governance framework, meaning it's not just the US deciding the rules. It's in line with his top intelligence official Chen simultaneously calling for global cooperation on risks, while avoiding allowing one country to monopolize AI power (a thinly veiled jab at the US). Beijing likely won't agree to a Washington-mandated pause, but if it wants to lead the world in the technology, it must be willing to work with other nations on the threats both sides can at least agree on. Doing so might not halt the race, but it's a baby step toward making it far less reckless.

The latest safety rallying cry in the US is the incredibly hollow phrase "pace the frontier," a jargony proposal from Anthropic PBC's Dario Amodei that garnered a rare agreement from OpenAI's Sam Altman and xAI's Elon Musk. A statement supporting this concept has accumulated more than a thousand signatories from nearly all of Silicon Valley's leading AI companies. But few seem to know what that actually means. If they're legitimately concerned about self-improving AI ending humanity, vague calls for slowing down is a remarkably feeble response.

Still, it would be in these firms' interest to spend less resources trying to build a machine God and more ensuring that their products aren't being abused. They can't unleash cyber weapons and then simultaneously sell defenses. Rather than a supposed "pacing" of AI development on their own terms, policymakers must demand maximum transparency on the risks they're seeing now, not disclosures months after an incident.

To his credit, Amodei himself supported this in his more than 3,800-word manifesto published after one of his employees quit and sounded the alarm that the people building AI believe it could "kill us all." Regardless of any commitments they make, Amodei said, "The public deserves to know what is going on." He's correct.

Beijing and Washington might not be able to agree on a lot when it comes to AI governance. But the WeChat episode is a sign that cross-border risk sharing is possible. AI has already changed the internet and privacy as we know it. Imagine a cyberattack that exposed every private communication you've ever sent and the havoc that would unleash across businesses and communities. Creating channels for researchers and companies to share what they're seeing and flag emerging threats will inherently tame some of the race dynamics and create an opening for the harder debates.

If the US and China can't cooperate on the dangers already in view, there's little hope they'll manage the ones AI's own builders warn could be the end of us all.

Catherine Thorbecke is a Bloomberg Opinion columnist covering Asia tech. Previously she was a tech reporter at CNN and ABC News.