What Trump gets wrong about US power
Plus: Brain spoons, super horses, climate spending woes, AI safety woes, a “transformative moment” for the Middle East, and more!
—Anthropic is offering up to $20,000 to the first person who “jailbreaks” its latest AI model by getting it to divulge things about making chemical weapons that it’s not supposed to divulge. We would remind any NZN reader whose prompt engineering skills empower them to snatch this prize what Peter Parker—aka Spiderman—was told by his uncle after getting superpowers via the bite of a radioactive spider: With great power comes great responsibility. We would also remind this reader that contributions to the Nonzero Foundation, which publishes NZN, are tax deductible. (See below for reason to worry that a very powerful new AI model released by OpenAI this week was rushed out, under competitive pressure, before being adequately tested.)
—Americans who worry that gene-splicing breakthroughs in another country could threaten US supremacy typically have China in mind. But if you’re worried about a master race of foreign horses subjugating American horses, the country to keep your eye on is Argentina, where, Reuters reports, a biotech firm has used CRISPR-Cas9 to produce at least five “polo super ponies.”
—Abdullah Ocalan—jailed leader of the PKK, a Kurdish militant organization—will soon make a “historic call” for an end to his group’s insurgency against the Turkish government, according to Turkey’s pro-Kurdish political party. Journalist Murtaza Hussain said such a move would be a “transformative moment in the Middle East,” with particularly positive impact in Syria, where the fledgling Turkish-backed government is struggling to find a modus vivendi with PKK-linked militias in the country’s northeast.
—The Trump administration warned over 1,100 workers at the Environmental Protection Agency that they could be fired at any time, the New York Times reports. Many of the employees now on the chopping block work on projects like the replacement of lead pipes and the remediation of toxic sites. (More on Trump’s shock-and-awe approach to environmental spending below.)
—The White House wants Ukraine to resume holding elections, which were suspended after Russia invaded and martial law went into effect, Reuters reports. In one scenario, Ukraine would commit to elections as part of a preliminary ceasefire, which could mean that a longer-term truce was negotiated by a Ukrainian president other than Volodymyr Zelensky.
—Google, which famously built its identity around the “don’t be evil” motto, has quietly dropped a pledge from its ethics guidelines to avoid developing AI for use in weapons and other technologies that “cause or are likely to cause overall harm.” The change comes as big tech draws closer to the US national security apparatus, and it happens to coincide with a disappointing quarterly earnings report for Google’s parent company, Alphabet.
Trump’s first volley of tariffs ended as suddenly as it began. After threatening to impose a 25 percent levy on all goods imported from Mexico and Canada, the president relented in mere days after extracting promises from each country to take further steps to fight drug trafficking and secure its border with the United States. “I am very pleased with this initial outcome,” Trump posted, noting that tariffs would be suspended for 30 days. Rep. Lisa McClain (R-Mich.) put it in more effusive terms: “The days of America getting walked all over—my friends, those days are gone.”
Some of these commitments, such as a promise from Mexico to send 10,000 soldiers to the border, may prove to be substantive. Others, like Canada’s appointment of a “fentanyl czar,” look more like theater. Many more of them appear to have been in the works already. Either way, though, the blowup is a useful reminder of how unwilling President Trump is to accept the limits of American power in the world today—a refusal that risks further eroding American power over the long term.
Whether Trump likes it or not, the unipolar moment—that brief post-Cold War period of true American primacy—is over, particularly in economic terms, and Washington’s leverage is an easy resource to squander. States like Mexico and Canada would very much like to maintain strong trade relations with the US, but they no longer have to do so. Economic giants like China and the European Union (as well as mid-sized states like India or Brazil) are more than willing to help fill any gaps left by the United States.
Trump can get modest short-term wins by threatening sudden economic shocks. But over the long term, countries will try to limit their exposure to a volatile trading partner. In fact, countries around the world have already started looking for ways to reroute commerce around the US. As the New York Times reported, the EU has in the last few months struck new trade deals with Argentina, Brazil, Paraguay, Uruguay, Mexico and Switzerland. And European officials are now mulling a trade rapprochement with China, which Trump also targeted with tariffs this week. Britain, for its part, recently joined the trans-Pacific partnership, a free trade zone made up of states in East Asia, Oceania and South America. As one trade expert told the Times, the world economy is now seeing “ever deepening trade relationships excluding the United States.”
For most of the post-World War II era, America has used its status as the dominant global economic power to shape the rules of trade and hem in its adversaries. That power is still huge, but preserving it depends on a certain level of restraint, especially given all the other powerful states in the game today. The more that Washington hits other countries with sanctions or tariffs, the more it encourages other states to find ways to cut their ties to the American economy. Trump’s indifference to this danger—coupled with a zero-sum view of international relations—is leading to a lot of moves that might look strong but often wind up leaving the country weaker. To put it another way, Washington’s economic sword, once the world’s sharpest, is growing dull through a thousand mindless blows.
A new study found that the quantity of microplastics in the average human brain has risen by over 50 percent since 2016, the Washington Post reports. Researchers told the Post that we don’t yet know if all that plastic is making us dumber or moodier or anything else (although the brains of dementia patients have been found to have higher levels of microplastics). All they know for sure is that there’s a lot of it—enough in each brain, they now believe, to make a standard plastic spoon. At this rate, our brains may soon house a full set of cutlery.
Two weeks ago, after the new Chinese AI DeepSeek-R1 freaked out both Silicon Valley and Wall Street with its combination of capability and affordability, OpenAI CEO Sam Altman said, basically: Bring it on! In a post on X, he wrote that “it’s invigorating to have a new competitor” and vowed to respond by accelerating the unveiling of new products: “We will pull up some releases.”
He wasn’t bluffing! This week OpenAI released Deep Research, a variant of its latest “reasoning” AI (o3) that has already turned heads with its ability to scour the web and produce sophisticated research reports. Highly regarded AI watcher Ethan Mollick, a professor at the Wharton School, tweeted, “I have been getting a steady stream of messages from very senior people in a variety of fields who have been, unsolicited, sharing their chats [that they had with Deep Research] and how much it is going to change their jobs. Never happened with other AI products.” The new model has “rough edges and hallucinates,” Mollick said, “but I haven't seen senior people as impressed by what AI can do, or as contemplative of what that means for them (and their junior employees) as now.”
DeepSeek isn’t the only company whose competitive pressure was reflected in OpenAI’s latest product launch. By naming the model Deep Research, Altman seemed to be encouraging comparison with Google’s Deep Research, which was released in December and has basically the same job description as its namesake—but hasn’t yet been hooked up to Google’s latest reasoning AI and is said to have weaker analytical skills than its new rival.
All of this raises a question:
For years AI safety advocates have warned that intense competition among AI companies could lead them to release models that hadn’t been adequately tested. Is that starting to happen? Did OpenAI rush out Deep Research without giving it proper scrutiny?
That question came up this week in a NonZero podcast conversation with Dan Hendrycks, head of the Center for AI Safety and a prominent figure in AI research. His answer wasn’t reassuring.
The full conversation with Hendrycks won’t be posted until the release of a paper he’s co-authored that was the focus of most of the discussion, but here is a (lightly edited) transcript of the part that dealt with the testing (or lack thereof) of the large language model that powers Deep Research: