Daniel Schwarcz

Fredrikson & Byron Professor, University of Minnesota Law School. Insurance law, regulation, & AI. Research available at

[Daniel Schwarcz @Dschwarcz]

Try to imagine a world in which we had to turn off all computer systems linked to the internet b/c of rogue AI agents. Financial , healthcare , transportation, and utility systems would be thrown into complete chaos, plausibly triggering broader societal collapse.

[Frank Luntz @FrankLuntz]

Here is a leaked top-secret plan for stopping AI if it tries to kill all humans:

[Daniel Schwarcz @Dschwarcz]

6/By redirecting efforts to develop frontier models into efforts to realize the enormous untapped value of the AI we already have, we may be able to reach a political compromise that involves reducing catastrophic risk without sacrificing AI’s promise.

[Daniel Schwarcz @Dschwarcz]

1/ I’m increasingly persuaded by calls to regulate frontier AI to slow its progress given the existential risks involved. But for a slowdown to be politically and economically feasible, it needs to come with an affirmative vision of what happens instead.

[Jacob Coxon @hilbertspaess]

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

[Daniel Schwarcz @Dschwarcz]

5/ This is politically essential. Many people and firms now expect AI to generate enormous social and economic benefits, and are building those expectations into investment and planning. A slowdown strategy without more will therefore face enormous resistance.

[Daniel Schwarcz @Dschwarcz]

1/ I’m increasingly persuaded by calls to regulate frontier AI to slow its pace, given the existential risks involved. But for a slowdown to be politically and economically feasible, it needs to come with an affirmative vision of what happens instead.

[Jacob Coxon @hilbertspaess]

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

[Arpit Gupta (in DC Sep-Dec 2026) @arpitrage]

There is debate over how Pangram works, so I did an experiment.

1) I start with the US Constitution (reassuringly, 100% human written).

2) I next asked for edits to spelling and grammar only. This is 1% AI written.

3) I then asked for clarity rewrite. Much more AI written

[Daniel Schwarcz @Dschwarcz]

@FrankPasquale @BrennerFissell @ProfArbel @OrinKerr @MeganTStevenson @TaxLawProf @AlexNunn @msmith750 @brianlfrye @HoffProf No one is obligated to read work that is bad or unhelpful. I think we're still in a place where the process you describe would produce such work, which would be quickly apparent to educated readers. But I do worry what happens when that's no longer so, which may be very soon.

[Daniel Schwarcz @Dschwarcz]

1/3 The conversation on AI use & legal scholarship has led me to draft a new AI disclosure to add to my work on a going forward basis. I'm curious whether critics of AI use will find this use of AI objectionable. If so, why is it OK to delegate these tasks to a human law student?

[Daniel Schwarcz @Dschwarcz]

@FrankPasquale @BrennerFissell @ProfArbel @OrinKerr @MeganTStevenson @TaxLawProf @AlexNunn @msmith750 @brianlfrye @HoffProf I don't think we should conflate the questions of (i) the quality of work produced with AI assistance, & (ii) the appropriateness of using AI for legal scholarship. Clearly one can use AI to produce slop, and that's what you seem to be contemplating here.

[Daniel Schwarcz @Dschwarcz]

@TimSchnabel I'm wrestling with this exact same thing. I thank RAs for helping me articles, but I don't specify the tasks they performed. So why should I do so with AI? Right now my answer is just that this seems to be the emereging norm, and I want to play by the rules. But should it be?

[Daniel Schwarcz @Dschwarcz]

@BrennerFissell @ProfArbel @OrinKerr @MeganTStevenson @TaxLawProf @AlexNunn @msmith750 @brianlfrye @FrankPasquale @HoffProf Yes, writing is often thinking and so using AI to edit creates risks. But so does over-relying on a human RA. The author’s job is to manage those risks. Whether they do so should be judged by the quality of the work, not by whether the process introduced some risk.

[Daniel Schwarcz @Dschwarcz]

cc: @OrinKerr @MeganTStevenson @ProfArbel @TaxLawProf @AlexNunn @msmith750 @brianlfrye @FrankPasquale @BrennerFissell @HoffProf

[Daniel Schwarcz @Dschwarcz]

3/3 My own experience is that using AI well in this targeted fashion substantially improves both the efficiency and the quality of my work, a result that is consistent with my overall research on how AI impacts legal work more generally. https://t.co/TfKJpP9cvt

[Daniel Schwarcz @Dschwarcz]

5/5 Institutional legitimacy matters. So does effective regulation. We should be capable of confronting both problems at once.

[Daniel Schwarcz @Dschwarcz]

1/5 I was honored to see my scholarship referenced in a Wall Street Journal op-ed this weekend: https://t.co/SiUAZZ588u

[Daniel Schwarcz @Dschwarcz]

4/5 But that doesn't mean that efforts to rein in private-equity backed insurers’ excesses are misguided. Those efforts are increasingly vital as allegations concerning the Walter insurance empire bring the risks of inadequate oversight into focus. https://t.co/2NGYrFr53H