Can Claude Write "The Situation"?
An experiment with replacing myself with AI—plus a new feature for paid subscribers
Good Afternoon:
(1) I recommend threats of violence.
(2) Easy. Don’t make X-rated work Zoom calls.
(3) No.
(4) No. You’re estranged. That means you don’t tell him things. But maybe tell your biological father that he you have an estranged brother and he has an estranged son.
Thursday on #DogShirtTV, the estimable Anastasiia Lapatina gave us Project Battery updates and explained how things are going in Kyiv now that the weather is warming up:
Friday on #DogShirtTV, the estimable Scott Shapiro came on to share his experiences with AI agents and some thoughts on the damage such agents can cause:
Recently On Lawfare
Compiled by the estimable Marissa Wang
In Case of Emergency: The Dubious Legality of Trump Allies’ Draft EO
Anna Bower and Molly Roberts explain that the Trump administration’s possible invocation of a national security emergency to justify sweeping federal control over U.S. elections would likely raise constitutional conflicts and fail under legal scrutiny.
Whether or not Trump is considering the precise recommendations the Post describes, the recent story is only the latest indication that he and the people around him view the declaration of a national emergency—related to national security—as a possible way to exert more control over elections. And while the contours of such a ploy remain unclear, there’s enough information available today to get a sense of what authorities the White House might invoke to try to pull it off.
One thing is for sure: Whatever Legal Argument (or plain old legal arguments) Trump may plan to present, it will be far from “irrefutable.”
Military AI as ‘Abnormal’ Technology
Scott Sullivan argues that policymakers should consider artificial intelligence (AI) in the military as an abnormal technology—diffusing unevenly amongst civilian and military sectors—because competitive incentives, externalized costs, and operational secrecy pushes military AI to develop faster than AI governance.
If military AI is an abnormal technology—defined by accelerated incentives, externalized costs, and epistemic opacity—then governance cannot simply mirror civilian models. It must instead anticipate structural pressures that erode traditional safeguards. This may require earlier-stage legal interventions than we’ve seen so far, anchored in the proposition that law of armed conflict rules do not merely regulate outputs at the point of strike, but, at least in high-stakes operations, necessitate design and development choices to reasonably ensure that the deployment of AI-enabled systems will be lawful by default.
The IEEPA Tariffs Are Gone. The Damage Is Done.
Stratos Pahis argues that the delay from the courts in stopping President Trump’s tariffs under the International Emergency Economic Powers Act has caused lasting harm to consumers, the U.S. economy, and U.S. credibility with allies.
The Supreme Court made the right decision. As a matter of law, the Court was correct to find that the International Emergency Economic Powers Act (IEEPA) does not confer tariffing power to the president. As a matter of policy, the Court was right to stop the president from imposing any tariffs on any country for whatever reason he wants.
Unfortunately, the president’s illegal tariffs have already caused damage that no court can undo.
Huawei Is Betting on the Future with HarmonyOS
Samm Sacks examines how Huawei is using its homegrown operating system, HarmonyOS, to integrate artificial intelligence (AI) agents across its devices and products, positioning the Chinese company as a potential global leader in agentic AI.
Recent updates to Harmony aim to integrate layers of China’s AI ecosystem that have traditionally been walled off and that have made AI agents less useful and appealing to users. Huawei executives have suggested that they intend for HarmonyOS to eventually rival Apple iOS and Google Android as the world’s third operating system. If Huawei succeeds, HarmonyOS may one day soon run on devices for global export—from cars to phones to fridges—with its AI-enabled assistant coordinating across all aspects of users’ lives. Chinese AI integration, in other words, will affect everyone everywhere.
Trump’s Cyber Strategy… Great, Amazing, The Best Yet
In the latest edition of the Seriously Risky Business cybersecurity newsletter, Tom Uren unpacks Trump’s newly released National Cyber Strategy, the misuse of privately developed cyber espionage kits by adversaries, and more.
President Donald Trump’s Cyber Strategy contains an ambitious array of worthwhile goals. The administration’s actions over the past year, however, directly undermine many of them, barring one. It raises the question: Can aggressive offensive cyber action compensate for lukewarm defensive efforts?
Podcasts
On Thursday’s Lawfare Daily, Matthew Olsen joins me to discuss the terrorist threat from Iran, the U.S.’s lack of preparedness to respond after firings at the FBI and the Department of Justice’s National Security Division, and the pending lapse of the FISA 702 program.
On Rational Security, Scott R. Anderson sits down with Troy Edwards, Kate Klonick, and Roberts to unpack Anthropic’s civil complaint and petition for appeal against the Department of Defense’s designation of the AI company as a supply chain risk, the threat of Iranian terrorism against the U.S., and the federal investigations into claims of election fraud during the 2020 presidential election.
On Friday’s Lawfare Daily, Charles Lister joins Scott R. Anderson to talk about the transitional government in Damascus, its relationship with the Syrian Democratic Forces, and the impending withdrawal of U.S. troops after more than a decade in the country.
On Scaling Laws, Kevin Frazier sits down with Tomicah Tillemann to discuss how to regulate emerging technology through policies that enable human agency and flourishing in an AI-powered world.
Videos
On Friday, March 13 at 4 pm ET, I sat down with Roger Parloff and Kate Klonick to discuss Anthropic’s legal challenges to the Defense Department’s supply chain risk designation, challenges to the legality of certain U.S. attorneys, litigation over the Trump administration’s immigration policies, and more.
The Situation: An Investigation Based on Vaporous, Speculative Malice
The Situation on Tuesday considered what Anthropic means by its red lines on military use Claude.
On Friday, James Boasberg—the chief judge of the U.S. District Court for the District of Columbia—found himself in another spat with the Trump administration over that little rule of law thingy.
The judge unsealed an opinion from a few days earlier, in which he had quashed a pair of subpoenas in the investigation of Federal Reserve Chair Jerome Powell.
In response, U.S. Attorney Jeanine Pirro gave an unhinged press conference in which she lambasted him as an “activist judge”—though she refrained from commenting on his weight—and swore a blood oath to file a motion for reconsideration and then appeal. She also said she was willing to see acquittals and willing to see grand juries reject her proposed indictments—which was refreshingly candid from a woman who has seen her fair share of both in her short tenure at the helm of an office that should see very few of the former and absolutely none of the latter. She also yelled a lot.
Remarkably, the government’s brief in the matter is scarcely less of a temper tantrum than Pirro’s press conference. It clocks in at a scant nine pages long—which is 38 pages shorter than the Fed’s motion to quash the subpoenas, to which it responds.
And Judge Boasberg is not exaggerating when he says that it contains “essentially zero evidence to suspect Chair Powell of a crime.” Indeed, it contains hardly more indication of what the government is even investigating. The document notes that there have been cost overruns in the Fed’s renovations. And it claims there are unspecified discrepancies in a congressional testimony given by Powell. But it doesn’t remotely suggest a basis for thinking that Powell might have lied in his testimony or that criminal conduct led to the cost overruns. It actually doesn’t even try.
The striking thing about the brief, and about Pirro’s press conference, in fact, is that neither seems remotely concerned with establishing that there is a predicate for a criminal investigation at all. In her press conference, Pirro waxed angry that Boasberg would presume to situate himself between the grand jury and the evidence it needs to decide whether a crime has been committed—or to satisfy itself that a crime has not been committed. And the brief does a fair amount of that too, although in slightly less extravagant terms.
But neither shows any awareness that investigative agencies aren’t supposed to initiate criminal investigations at all without an appropriate evidentiary predicate. It’s unclear from the record in this case so far what investigative agency is actually conducting the probe. But normally, to open an investigation in the first place, the agency has to have specific articulable facts indicating that a crime has been or is being committed.
The U.S. attorney’s office, both in court and in public, has said nothing that seems like a legitimate predicate for an investigation. Yes, Powell gave testimony and—according to Pirro—something he said (but don’t ask what) may be inconsistent with something in the public record. And yes, there were cost overruns in the Fed’s renovation, but that is not a crime.
Pirro on Friday mentioned a statute that criminalizes false statements to Congress at her press conference yesterday, but she didn’t mention anything Powell said that might have been a false statement—let alone an intentionally false statement on a material point of fact. If the government has specific articulable facts indicating that a crime has been committed, it’s keeping very quiet about them. It is putting forward only a scant few very non-specific facts, and if it has more, it is choosing not to articulate them.
Let me lay my cards on the table: I don’t believe there is a proper predicate for this investigation.
I think the subpoena issued on the most vaporous of speculative malice. Everything that has come out so far affirms Lawfare’s Roger Parloff’s instinct on the matter from several weeks ago:
[T]he Trump administration, when publicly embarrassed, tends to vindictively double down. If it follows that formula, Pirro or Trump’s other allies will scour the earth searching for new dirt on Powell. Maybe that was the whole reason for the grand jury subpoenas. Maybe Pirro wants to look for purported discrepancies between what Powell said in his testimony and what’s occurring on the ground today at the renovation construction site. Perhaps she’ll find a vegetated roof that’s five feet above ground, and not truly at ground level. Maybe she’ll find some new piece of white marble that is not merely replacing crumbling, old white marble. Maybe she’ll even find some beehives.
Here is what’s not going to happen: First, Judge Boasberg is most unlikely to reconsider his ruling, nor should he—at least not without some hint that there is an evidentiary basis for this subpoena and that is not a scouring of the earth for dirt on Powell. Second, there are very few panels of the D.C. Circuit Court of Appeals that would seriously contemplate reversing Judge Boasberg on this record, though there may be one or two.
That means the question boils down to this: Do Pirro and Solicitor General John Sauer want to take this rather barren record, a record strongly suggestive of malevolently political use of the criminal process, to the Supreme Court?
Normally, such a thing would be unthinkable. Then again, normally, it would be unthinkable to issue such subpoenas in the first place, much less to litigate with them with nine pages of bluster that lacks so much as the remotest evidentiary justification. And let’s face it: Putting this matter in front of the Supreme Court is no crazier than, say, announcing you are withdrawing the appeals of adverse rulings in the law firms’ cases—only to backtrack and then tell the courts you are proceeding with those appeals after all.
In other words, we may end up establishing appellate precedent, at one level or another, on a question of almost mind-boggling stupidity: Is it proper for a court to quash a grand jury subpoena where there is an overwhelming prima facie case of political motivation on the part of the prosecutor and no clear allegation of wrongdoing on the part of any subject, indeed, where there is no clear predicate for an investigation at all?
The Situation continues tomorrow.
Today’s #BeastOfTheDay is, once again, the baby elephant, seen here being thoroughly offended by a human refusing to pay attention to it:
Please also enjoy this still of today’s Beast as it formulates its plan of attack:
In honor of today’s Beast, check out the Sheldrick Wildlife Trust, where today’s Beast—along with more than three hundred other baby elephants—was raised.
In further honor of today’s Beast, find someone filming a video and grab their face. You can tell them it’s a Tiktok trend or something.
A New Feature For Paid Subscribers
With this dog shirt, I am rolling out a new feature for paid subscribers: a weekly roundup of judicial opinions, litigation filings and other documents I found interesting over the course of the week. I plan to collect material over the course of every week and then assemble it in one weekend dog shirt. It’s designed to be useful and interesting. Please let me know what you think of it and how it can be more so.
Can I Replace Myself With Claude?
Okay, folks, let’s face the big question: Am I obsolete? Can an AI do my job? Or, to be more precise, how much of my job can an AI do—and at what level of quality relative to the quality of my own all-too-human output?
I want to face this question with unstinting honesty and transparency. If I’ve become a useless meat sack, I’m going to face that fact head on and without denial.
Yesterday, I wrote my regular “The Situation” column for Lawfare. I did it without any AI assistance of any kind. That means I read all the documents. I composed all the sentences myself. I put it through human editing. What you see on Lawfare—and above—is a purely human writing process.
Having written it myself, I then had a brief exchange with Claude in which I asked the Anthropic AI to write my column for me. I did this with no intention whatsoever of publishing the column it wrote as my own. Rather, I did it as an experiment to see how much of my work product is delegable to an AI. What percentage of Wittes-level production can you buy with an Anthropic subscription?
My first effort to get Claude to write my column was a total failure, because Claude didn’t wait for my instructions as to what I wanted it to say before spewing out a rather windy column. In response to my first prompt, it produced a whole lot of words about the tariffs decision—words I would never write.
“[Y]ou jumped ahead,” I said in response. “I was actually going to give you specific instructions about the column I wanted you to try to draft.”
“You’re right, I got ahead of myself!” Claude responded. “Please go ahead — what are your instructions for the column you’d like me to draft?”
This time, I gave a very specific prompt based on the column I had written. I intentionally kept these instructions vague, as I wanted to see how Claude would do with a high-level task and not too much guidance. Would it produce a Wittes-level column? Something utterly useless? Or something at the level of a good research assistant?
Here’s the prompt:
Keep reading with a 7-day free trial
Subscribe to Dog Shirt Daily to keep reading this post and get 7 days of free access to the full post archives.
 | A guest post by
|