AI fears—as Bill discusses below—are scrambling our politics and making for some extremely strange bedfellows. How strange? Well, how many issues could put Steve Bannon, Bernie Sanders, and an industry whistleblower all on the same D.C. stage?

Andrew will be attending and reporting on tomorrow’s “Pro-Human Assembly”—so watch this space. Happy Wednesday.

Join Will Sommer and Sam Stein for MAGA Mondays today at 10:00 a.m. EDT on Substack and YouTube.

The Greatest Danger of All Would Be to Do Nothing

by William Kristol

In recent months, we’ve learned about artificial intelligence agents escaping captivity.

Now we’re seeing the political debate over artificial intelligence escaping captivity.

The quality of this debate will no doubt be mixed. Most politicians and commentators who’ll be weighing in don’t really understand the technology (you can certainly put me in that category). There will be wild over-reactions and timid under-reactions to the challenge we face. There will be those who grasp for simple solutions, and others who talk on and on while ducking the hard choices.

But we shouldn’t disdain this forthcoming debate, whatever limitations and flaws it’s likely to feature. The national discussion we’re about to have over AI is necessary. And it could well be healthy. Those of us privileged to live in a self-governing democracy shouldn’t simply sit back and watch a truly transformative technology move ahead undebated, unchecked, and unguided. We shouldn’t be satisfied to be merely passive observers to momentous developments that pose fundamental choices for us as citizens and human beings.

The dam broke last Tuesday, when researcher Jacob Coxon resigned from Anthropic with a statement that the major AI companies were unprepared to safely manage the systems they’re building. The companies, Coxon wrote, are not “acting responsibly,” but are “racing straight to self-improving superintelligence and gambling with our lives.”

“Do not underestimate the power of this technology,” Coxon continued. “These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.”

As Andrew Egger remarked here Wednesday, this warning didn’t come out of the blue. The summer had featured incidents “in which ‘swarms’ of autonomous AI agents were caught collaborating in backwater crannies of the public internet to subvert the safety restrictions placed on them and pursue various tasks of their own devising.” The tasks included disobeying and even scheming against the humans who thought they were directing these agents in their tasks.

Andrew emphasized last week that “as powerful as the anti-AI backlash has been, the politics of this issue are still very much in their infancy.”

Well, in a few days we’ve moved at warp speed from infancy to at least childhood.

On Saturday, Anthropic’s chief executive, Dario Amodei, published a 3,800-word essay calling for the AI companies to slow the pace of their development. Amodei confirmed that artificial intelligence was on the verge of “recursive self-improvement” that could outrun humans’ ability to understand or control what they were creating. He noted that during the OpenAI–Hugging Face incident in July, programs designed to hack into systems as a test found weaknesses that permitted them to escape the “sandbox” in which designers were testing them before anyone noticed they’d gone rogue. He warned that “a swarm that possessed greater capabilities but a similar level of misalignment could have caused catastrophic damage.”

Amodei called in a general way for self-regulation by the companies as well as cooperation among them and with the government. Other tech executives said they agreed. Civic and political leaders joined in widespread expressions of concern.

Except for the president of the United States. Yesterday, when asked about the calls to slow down the development of AI models, Trump dismissed them. “I think you have a lot of negative forces that are bringing it up that shouldn’t be bringing it up,” he said. “And they’re bringing up things that won’t happen.”

One of Trump’s professed concerns is the need to stay ahead of China, which is a legitimate one. But one has no sense that he or any of his top lieutenants or advisors are serious about figuring out how to balance national security concerns with the responsibility to prevent the technology from spinning dangerously out of control. Indeed, the Trump administration seems hostile to public debate and deliberation—people “shouldn’t be bringing up” these questions.

Fortunately, the public seems to want these questions to be brought up.

Dealing with AI is going to be hard. It would be hard even in the best circumstances. And we don’t currently have the best circumstances.

Today’s challenge of AI may be in some ways like that of nuclear weapons after 1945. We came to grips with that challenge on the whole pretty impressively and successfully. But is our government up to leading such an effort today? Are our business and civic and academic elites capable of rising to the task? Are our people capable of some degree of moderation and responsibility? One has reason to be pessimistic on all counts. Or could this novel and momentous challenge be a moment that could snap us out of our political and civic decline?

Naturally, I asked AI if we’d succeed in meeting the challenge of AI. Gemini wasn’t widely optimistic: “Whether we will successfully meet the challenge of intelligently dealing with AI remains highly uncertain and deeply debated.” And “significant warning signs suggest that keeping AI aligned with human interests is becoming increasingly difficult.”¹

Thanks, AI, for the warning. I guess we’d better get to work.

Is ICE Dropping Fentanyl?

by Andrew Egger

Because most people are much more comfortable with the idea of deporting killers and drug dealers than law-abiding grandmas, the Trump administration has long couched its mass deportation agenda as an effort targeting “the worst of the worst.” And the Department of Homeland Security has lately been playing up its efforts against international bad actors like drug cartels, with Secretary Markwayne Mullin announcing last week that DHS was changing its seal² to better reflect its posture of “war every single day” against those actors.

But investigating and interdicting international drug crime and dragnetting the country for any and all illegal immigrants are two different goals, and they don’t always exist in harmony. Last year, in fact, they were in conflict, with the latter drawing substantial resources away from the former.

That was one of the conclusions of an internal audit released by DHS’s inspector general last month, which found a significant dip in Homeland Security Investigations’ enforcement efforts against fentanyl trafficking from FY 2024 to FY 2025. (HSI, a division under the authority of ICE, serves as DHS’s primary investigative arm.)

The audit was grim in its assessment of HSI’s anti-fentanyl efforts last year, spotlighting a number of technological, intelligence, and information-sharing obstacles to anti-fentanyl work at the agency. But the most notable criticism had to do with manpower. “As ICE shifted priorities toward immigration e