Tessera
Folgen

John Kastellec

@jkastellec.bsky.social

Professor of Politics at Princeton University @Princeton . http://www.makingthesupremecourt.com

214 Posts6213 Follower383 Folgt
Kieran Healy@kjhealy.co · 14h ago

Well this is interesting in the ongoing Authentication/Authenticity arms race. Anthropic says it will start watermarking all text Claude generates. They are a little vague but seem to mean some sort of token-biasing steganographic signal, not a watermark you can strip out by just retyping the text.

All Collections > Privacy and legal > How Claude marks Al-generated content How Claude marks AI-generated content Updated today • Copy for LLM Anthropic has signed the EU AI Act's Article 50(2) Code of Practice on Transparency of Al-Generated Content, as a provider of both generative Al models and generative Al systems. This article describes how we're planning to put those commitments into practice, how marking works, and what its limitations are. We'll update this article and publish more detailed technical guidance as it becomes available. Anthropic's commitments under the EU AI Act's Code of Practice on Transparency of Al-Generated Content What our marking commitments mean for Claude: • New models will mark Al-generated content from day one. Claude models launched in the EU on or after August 2, 2026 will support machine-readable marking at launch. Generated text will carry embedded watermarks, and generated files will include digitally signed provenance metadata where supported. • Marking works everywhere you use Claude. Marks will apply to output from supported Claude models across Claude Platform (API), Claude, Claude Code, Claude Cowork, and Claude Tag, and wherever Claude is offered, worldwide. Some platforms or features may not support certain marking types. • We'll help you detect Claude's marks. We'll support users and other third parties to detect Claude's marks, as the Code requires, and we'll share details in forthcoming documentation. • Existing models are in progress. The law includes a transition period for Anthropic models launched before August 2, 2026, and we're working to add marking support for those models as well. More details about our marking plans are below.How Claude marks content Claude uses two complementary techniques to mark content generated and processed by Claude: (1) watermarks embedded in text, and (2) signed provenance metadata attached to files. 1. Embedded watermarks in text When a supported Claude model generates text, it weaves an imperceptible watermark directly into the text itself. You won't see it, and it doesn't change the meaning, quality, or readability of Claude's response. Because the watermark is part of the text, it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing. Watermarking will be applied at the model level, which means it will be present no matter which Claude product or surface the text comes from. 2. Signed provenance metadata When Claude generates a supported file type, such as a svg, png, or jpg, it will attach signed provenance metadata. This metadata follows the Coalition for Content Provenance and Authenticity (C2PA) open standard, which is used across the industry to record information about content provenance. If a signed metadata label is present, it signals that a file was processed by Claude and lets you detect whether the file has been tampered with. Detecting Claude's marks We're also working to enable users and other third parties to detect Claude's embedded watermarks and provenance metadata. Detection checks whether a piece of text or a file carries a supported Claude mark. If a supported mark is found, it indicates that the content may have been processed by Claude. We'll share details on detection mechanisms in forthcoming technical documentation.Limitations Machine-readable marks provide important signals about content, but it's worth understanding their limitations across all content types. • A detected mark provides a signal that content was processed by Claude, but is not fully conclusive. Detecting a Claude mark tells you that the content may have been processed by Claude. It does not, on its own, confirm the full provenance of the content. For example: • Claude may not be the original author. People often use Claude to proofread, translate, summarize, or convert files. The output can carry a Claude mark even if the underlying ideas, text, or data originated from another source; • The content may have changed after Claude processed it. Marked content may be modified, excerpted, or combined with other material after Claude processed it. • Lack of a detected mark doesn't mean the content wasn't Al-generated or processed. Content generated by Claude may not carry a detectable mark if, for example: • It was generated by a model released before marking was supported; • The text has been heavily edited, paraphrased, translated, or mixed into other writing; • The passage is very short, leaving too little text for a reliable signal; • A file's metadata was stripped through format conversion, re-saving, screenshots, or other means; • It was produced through a platform, feature, or file type where a particular marking type wasn't supported. If you build with Claude If you deploy Claude in your own product, you should independently assess what Article 50 requires of your products and services. Consistent with our commitments under the EU Code, our goal is to support you in meeting your own transparency obligations, and we'll share technical guidance on our marking and detection approach as it becomes available.
Scott Lemieux@lemieuxlgm.bsky.social · 3d ago

In the words of an eminent analytical philosopher of the contemporary academic scene . . . www.lawyersgunsmoneyblog.com/2026/08/in-t...

In the words of an eminent analytical philosopher of the contemporary academic scene . . . - Lawyers, Guns & Money

. . . I’m losing my motherfucking mind here. Jesse Singal on the apparently egregious confabulations of ex-Cambridge professor Jason Arday: If interest in this story peters out, that would be unfortun...

lawyersgunsmoneyblog.com

Gabe Lenz@gabelenz.bsky.social · 4d ago

Codex has taken the pain out of finishing a data-heavy book. Here’s a paragraph about it. Being able to talk to your book and data files feels Star Trek-like. Link to my agent instructions below. ...

Bild
John Kastellec@jkastellec.bsky.social · 4d ago

I understand the desire and incentives for emphasizing flashy results from working papers, but, flawed as it is, peer review in top journals will often catch obvious methodological issues, especially when it comes to identification concerns and measurement issues.

Steven White@notstevenwhite.bsky.social · 4d ago

I worry the implications of this are getting blown out of proportion. Do @adambonica.bsky.social and @jakemgrumbach.bsky.social really not think, for example, Hong is less likely to win WI-Gov than Crowley? Or that AES performs as well as Stevens in the general (margins, not win vs loss)?

Zeeshan Aleem@zeeshanaleem.bsky.social · 4d ago

Now a new working paper by @adambonica.bsky.social & @jakemgrumbach.bsky.social studying millions of registered voters between 2016 and 2024 and found something remarkable: the penalty for extremism has vanished. www.dropbox.com/scl/fi/mjih7...

John Kastellec@jkastellec.bsky.social · 4d ago

I don't understand how one can note (correctly) in the paper that the 2.5 million vote claim is an "accounting benchmark rather than a counterfactual election result" and at the same time claim that the IRA "moved" 2.5 million votes, which is clearly a causal claim.

Leah Stokes@leahstokes.bsky.social · 6d ago

The conventional wisdom is the Democrats' big climate law failed to deliver votes and then died. That's wrong. My new study finds its factories moved 2.5 million votes—more than Trump's margin. And the law's still alive. My latest in @theatlantic.com. Gift link! www.theatlantic.com/ideas/2026/0...

Ryan Enos@ryanenos.bsky.social · 7d ago

Don't posit a theory that is only interesting because it implies X causes variation in Y, test for a statistical relationship between X and Y, but then say you are not interpreting causally. Of course you are interpreting it causally. That was the whole point of what you just did.

Daniel de Kadt@dandekadt.bsky.social · 2w ago

Happy to recommend that people read this, but would strongly encourage that reading to take place in conjunction with my original comment. A number of claims in this now published response are outright false or misleading, and a careful reading of the two in parallel will make that very clear.

Cambridge University Press Political Science & IR@cambup-polsci.cambridge.org · 2w ago

#OpenAccess from @apsrjournal.bsky.social - Still Instrumentally Inclusive - https://cup.org/4bhrhsp "Do individuals in Western democracies shift their views on LGBT+ inclusion when exposed to opposition from Muslim out-groups?" - @turnbulldugarte.com & @bertous.bsky.social #FirstView

Banner with the hashtag #OpenAccess on a green background and the text "American Political Science Review" on a blue background below.
Michael Clemens@mclem.org · 2w ago

The US Administration just restricted the duration of student visas. What's the problem? Now political appointees have full discretion to block most international students from staying after graduation. They've promised to use their new power to do just that. My new @piie.com post explains—>

A new US rule to restrict student visas will hurt the US economy

The Trump administration issued a final rule in July that gives it discretion to cut short the course of study of most international students at US universities. It will likely use this new power to b...

piie.com

Jake Anbinder@jakeanbinder.bsky.social · 3w ago

Been reading a lot about the DLC and more generally attempts to reorient Dems in the '80s and '90s and you can (and maybe should) disagree with the policy outcomes but you have to admire the level of coherence about strategy and vision which is not present today in any wing of the party

Eric Columbus@ericcolumbus.bsky.social · 3w ago

Here’s an idea: as part of a deal with Iran to end the war, let’s sell them weapons and use the proceeds to fund armed opposition to Daniel Ortega. Have we ever tried that?

The Washington Post@washingtonpost.com · 3w ago

Nicaragua’s President Daniel Ortega said his country won’t be holding elections, a visible effort to extend his nearly 20-year rule and dashing hopes by the opposition that he might heed their demands for a democratic vote. https://wapo.st/4gPZPp3

David Schraub@schraubd.bsky.social · 3w ago

The number of nominally Democratic commentators who want to effectively give the GOP another Senate vote -- who are *furious* at Beshear for NOT giving the GOP that vote -- bc something something "fighting" is mind-blowing. It is the apex of a truly toxic brainrot that is going to destroy us all.

Adam Cohen (My Personal Views Only)@axidentaliberal.bsky.social · 3w ago

It’s been over a week since @govandybeshear.bsky.social called on Mitch McConnell to disclose the state of his health There has been no credible, transparent response Beshear should call a #KYSpecialElection Anybody who has a problem with that should talk to Supreme Court Justice Merrick Garland