iiinova — where your digital self keeps evolving

Link to: https://x.com/bzamayo/status/2108498107184934914

I like this idea of “proof of human” but of course the next step will be having AI generate it. daringfireball.net/linked/2026/...

daringfireball.net

‘Proof-of-Human’

020

hilzoy @hilzoy.bsky.social · 10/10/2026

That depends on the jurisdiction's rules for statutory interpretation and on the statute in question. And, FWIW, computer-misuse laws like the federal CFAA were written to deal with viruses, worms, and other indirect ways of using computers and have been repeateldy applied in cases involving them.

000

Reposted by James Grimmelmann

It’s worth clarifying what’s wrong with this trial. Guinea-Bissau does not presently vaccinate newborns for HepB, so this study does not withhold a vaccine they would otherwise have gotten, or expose infants to a disease they would otherwise have been protected against.

39735

Zachary Schrag @zacharyschrag.bsky.social · 10/10/2026

No. Criminal liability is mostly based on statutes, rather than precedents. The big issue is the defendant's state of mind, not whether something like this has happened before. Civil liability depends on the causal chain, which can break if it is too long (10,000 agents).

100

Reposted by James Grimmelmann

Did I abuse Claude by showing it an 1844 newspaper?

Transcript from a Claude session.

Historian:
why am i getting this error: This response was blocked by content filtering. Rephrase your message or try again.

Claude:
Investigating the likely cause of a content filtering error.
I can’t see that error or what triggered it, so this is my best guess rather than a diagnosis: an automated filter checks responses as they’re generated, and this page is a hard case for it. It’s a nativist paper’s account of the Kensington riots, full of graphic descriptions of shootings and beatings and hostile language about Irish Catholics. A verbatim transcription reproduces all of that, and the filter can’t tell a historical source from new writing.

3215

It is darkly funny that the alignment-based paradigm within which the AI industry operates was designed to deal with AGI risks but seems much more suited to the mundane ones, while the most effective safeguard against the actual AGI risks seems to be mundane cybersecurity discipline.

060

“An unreleased, non-frontier research model”

“You do get how that’s worse, right?” meme

0110

"New" goals? The mathematical community has been having these converations for decades. Donald MacKenzie's _Mechanizing Proof_ is a good guide to how mathematics reacted and adapted to previous advances in automated provers.

120

This is going to end with AI company employees going to prison. Their lawyers are either incompetent or impotent.

000

They were still running these tests through SEPTEMBER? Did they learn nothing from the hacks in JULY?

The company discovered the error on Sept. 28. As a result, Gripp said, Anthropic terminated testing for that model and added safeguards for further trials.

101

They were testing *what* on the public Internet? This is getting into CFAA intentional access without authorization territory. This is how you commit actual crimes while doing AI testing.

The email was flagged as spam and remained in that folder until Wednesday, when Anthropic alerted police to the incident, authorities said. No city or police data was accessed as a result of the interaction, Gripp said.

153

But fortunately Anthropic’s defective computer was cancelled out by the police department’s faulty computer.

120

Ruha Benjamin @ruha9.bsky.social · 09/10/2026

This wasn’t Anthropic’s systems flagging a user prompt for investigation. This was a system just making up a tip.

The model gave information purporting to come from someone with information about an unsolved homicide, he said.

111

Reposted by James Grimmelmann

I’m in DC scheduled to give this talk today. But the organizers flagged slides in my deck that mention racism, structural gaslighting, New Jim Code and other themes that are off limits in the Land of the Free.

APA Women’s Leadership Institute event flyer with “keynote speaker” crossed out with a red X

729081215

If you click through to the article, you will learn as much about it as I know, and I doubt that anyone now living knows more.

020

Compare this to Lewis Lapham’s account of his CIA interview in 1957. www.laphamsquarterly.org/spies/open-i...

Prepared for nothing less, I had spent the days prior to the interview reading about Lenin’s train and Stalin’s prisons, the width of the Fulda Gap, the depth of the Black Sea. None of the study was called for. Instead of being asked about the treaties of Brest-Litovsk or the October Revolution, I was asked three questions bearing on my social qualifications for admission into what the young men at the far end of the table clearly regarded as the best fraternity on the campus of the free world:

1. When standing on the thirteenth tee at the National Golf Links in Southampton, which club does one take from the bag?
2. On final approach under sail into Hay Harbor on Fishers Island, what is the direction (at dusk in late August) of the prevailing wind?
3. Does Muffy Hamilton wear a slip?

151

Oh, yes, and sorry that I should have been clearer. It's better that they ask than cut back on the work; the bad thing is that they have to ask. Something can be the best response to an ugly fundraising environemnt, and still distort the mission.

210

BeijingPalmer @beijingpalmer.bsky.social · 08/10/2026

Part of this conversation should be about the high barriers—both technical and financial—that the ALI puts around access to its projects. Compare, for example, the difficulty of getting the text of a Restatement to the ease of getting the text of a Uniform Act.

030

Reposted by James Grimmelmann

extremely funny to draft your reports to your superior in a foreign intelligence or propaganda agency in a US controlled company, but I'd bet it's happening *a lot*. People are lazy! openai.com/index/disrup...

openai.com

Disrupting AI-enabled “false front” operations

OpenAI disrupted two AI-enabled influence operations that used false-front journalists and a think tank to spread geopolitical messaging.

217528

Reposted by James Grimmelmann

Vituperative Erb @vituperativeerb.bsky.social · 07/10/2026

I say again it’s sort of amazing the libs are getting both Great Books and the Protestant Work Ethic in the national divorce

2073991

Reposted by James Grimmelmann

Ryan Cordell @ryancordell.org · 08/10/2026

My child arrived just the other day He came to the world in an optimal way But there was code to vibe in Gemini So I trained a model what I might say And anytime he needed just to hear me speak I’d say “hey, please read like me, bot” “Bot, please read like me”

25415

Mor Naaman @informor.bsky.social · 08/10/2026

I’m glad to see tech workers with money support EFF in the same way I’m usually glad to see business leaders with money support universities, but putting this on the EFF blog feels like the university president saying “remember to give us your appreciated stock when you die” at graduation.

100

Reposted by James Grimmelmann

🆕 📝 New paper officially out in @nathumbehav.nature.com! In 2023, Change.org built an AI-assisted writing tool into their platform. We offer causal evidence that access to the AI tool changed petition text (word count, language)... but may have actually hurt outcomes (signatures, comments).

14318

Anil Dash @anildash.com · 08/10/2026

I for one think that it is a bad thing that the “leading nonprofit defending digital privacy, freedom, and innovation” is dependent enough on donations from the 1% of the U.S. population wealthy enough to have a donor-advised fund to post something like this.

460

Reposted by James Grimmelmann

What I *really* want an interviewer to do with these desperate sweaty CEOs is say, "Now that Trump has said he wants you to call it 'Trump Intelligence' instead of 'Super Intelligence', are you planning to update your marketing?" and then don't reveal that you made it up until after they say yes.

1001014177

Federal wire fraud requires (1) a falsehood, (2) that is material (i.e., matters to the person deceived), (3) made with the intent to deceive, and (4) the receipt of money or property as a result. Your laptop in the car example doesn't have (3) or (4).

010

No, but having bots stream your own songs, and lying to Spotify that they are actual human users is. That’s the fraud.

110

Just look at this fucking AI slop-bot reply.

010

🦇🎃💀 Riana-mator 💀🎃🦇 @riana.bsky.social · 08/10/2026

The application of this principle to the law-school hiring process and the American Association of Law Schools is left as an exercise for the reader.

24811

Reposted by James Grimmelmann

No it fucking didn't! Did anybody other than me actually read the fucking opinion? I feel like I'm taking crazy pills. (this is a guest column in The Hill which I will not dignify with a link, whose author, a legislator who otherwise sounds like someone who'd deserve my respect, should know better)

That’s got to stop now. A federal appeals court recently agreed but ruled that the current state of the law prevents law enforcement from prosecuting people who use AI image generators to create what is often called “synthetic child pornography.”

These materials can be lifelike depictions of toddlers engaged in sex acts with adults, or anything else you can imagine. But the court ruled that because the images don’t depict actual sexual abuse crimes against real children, they are free speech protected by the First Amendment.

2434