Words I did not enjoy reading this week, from a leading artificial intelligence company: “During our testing, we found that Mythos Preview is capable of identifying and then exploiting zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so.”
In plain English, Anthropic says its newest AI model has found security holes in the major systems that power … well, almost everything. Amateurs with modest coding expertise could conceivably exploit these holes to hack and crack a frightening chunk of the nation’s digital infrastructure. Pulling off such a feat might be expensive in terms of computing power, but a lot of governments and non-state actors have both money and skilled hackers to spare.
Maybe you suspect more AI hype, designed to goose the firm’s valuation. If so, note how many major companies are taking it seriously. Instead of releasing Claude Mythos Preview to the public, Anthropic is working with a consortium of key players such as Apple, Google and Microsoft to patch these holes as soon as possible. That’s a strong signal that the problem is real. So it’s sobering to imagine what might have happened if a less responsible company or government had gotten to this model first — because eventually others will get there. Anthropic may be leading the pack, but the pack is close behind.
Some will see this as more reason to ban AI before it steals our passwords and our jobs. Unfortunately, that won’t work, as this week’s events demonstrate, because the technology is out there, and if the United States doesn’t develop it, someone else will.
What if the Mythos breakthrough had occurred at a Chinese firm? Any such company of sufficient importance is effectively controlled by the Chinese Communist Party, which thinks AI is very important — so critical that the government recently barred two AI founders from leaving the country after Meta bought their start-up. If a Chinese AI developer had suddenly uncovered a wealth of security vulnerabilities, would that firm have been allowed to warn the world while helping propagate patches? Or would the exploits have been handed over to China’s extensive cyberoffensive operations?
The obvious rejoinder is bilateral talks are needed to enforce a worldwide pause. That’s an appealing but unworkable solution. It would amount to a major arms control negotiation, which can take years, if not decades, while AI develops new capabilities practically every month. Even if the diplomatic process is sped up, treaties are binding only as long as both parties agree to be bound.
America has a head start in the AI race because of its significant edge over China in what the industry calls “compute” (read: advanced chips), thanks in part to export controls. But China is way ahead of the U.S. in electricity generation and transmission, the other building block of advanced AI. In just the past four years, China has added an entire U.S. grid’s worth of capacity.
How, then, could the U.S. trust China to keep any deal? It might better suit China’s interest to agree to a pause now, while it races to catch up in computing power — and then withdraw as soon as it pulls even. That could effectively cement long-term dominance by China because it is likely to erase the compute gap long before the U.S. fixes its legal and regulatory barriers to infrastructure development.
All of which points to an uncomfortable conclusion: America can’t stop what’s coming. Instead of debating a data center moratorium, it must adapt.
Every U.S. company should probe its security, and so should every user, because any hacker who has been quietly sitting on a vulnerability in hopes of using it later now has reason to go for the gusto before the hole is patched. You should back up data and photos you’ve stored in the cloud, set up unique passwords for every online service you care about and secure critical logins with two-factor authentication.
Meanwhile, Democrats and Republicans need to bring the same passion to building out the nation’s electric grid that they brought to prosecuting culture wars. The country cannot afford protracted arguments over whether it needs more natural gas or more renewables, because the answer is yes: It needs an “all of the above” strategy that treats electricity abundance as a matter of personal and national defense rather than a political football.
The U.S. government’s technical capabilities need radical upgrading. We don’t yet know what AI governance should look like (imagine trying to write comprehensive modern roadway regulations in 1901!). So enacting strong preemptive rules now would more likely strangle development than advance it responsibly. But it’s clear the government will play a major role eventually, and that it is currently ill-equipped to do so. It’s not just Defense Secretary Pete Hegseth’s recent ham-fisted attempt to destroy Anthropic, which now looks particularly idiotic. The government’s hiring processes, salary bands, procurement rules and bureaucratic red tape make it impossible for the civil service to match private sector expertise.
Closing that gap demands major changes to nearly every facet of government operations. It will also require federal preemption of state and local regulation. State governments have even less technical capacity than the feds, making them ill-suited to manage fast-moving technology. More importantly, juggling 50 sets of competing — and possibly contradictory — rules governing development is not tenable. Many deployment questions, such as limiting access by children and procurement rules, should be left in local hands. But larger issues will need to be decided at the federal level, because maintaining America’s AI lead is critical to the nation’s economy and security.
Until now, most people have been thinking of AI policy primarily in domestic terms, as a question of how the U.S. will let it develop, or whether the country will allow it to develop at all. That narrow frame blinds us to the most important question: Will the nation’s future be governed by American values and American institutions, or by the CCP? If we get that question wrong, the others won’t matter much.
The post A frightening chunk of our digital world has become more vulnerable appeared first on Washington Post.
