Your Claude Diary Has a Reader -- AI Brief October 6

Good day %%first_name%%. A Florida woman’s Claude diary reached the police, ChatGPT is signing real cartoonists’ names to cartoons they never drew, and a16z finds that AI’s top one percent of payers spend $903 a month. Also on the menu: a free tool that evicts Apple Intelligence from your Mac, and Walter Isaacson calling on Ada Lovelace. Let’s get into it.

Her Claude Diary Ended Up With Police TechSpot

* What happened: A Florida woman, Carli Michelle Heller of Bonita Springs, faces a felony charge after allegedly writing in Claude on September 26 that she planned to “shoot up” the Sheriff’s office. She told investigators she uses the chatbot like a diary. Per the arrest report TechSpot cites, Claude’s safety systems flagged the entry, a human reviewer judged it a credible threat, and Anthropic reported it to law enforcement. The charge is a written threat of violence, a second-degree felony under Florida law.

* Why it matters: A chatbot isn’t a locked notebook. The company says it may share user information in limited emergencies when it believes disclosure is needed to prevent death or serious physical injury, and the flag here came from automated screening before a person made the call. Keep that in mind the next time you type something into a chat box you’d only say out loud to yourself.

* What everyone’s saying: The Hacker News thread splits two ways. One camp has sympathy for Anthropic, since OpenAI took heat after it didn’t report a shooter in a similar situation, and TechSpot notes British Columbia is suing OpenAI over exactly that. The other asks whether typing into a chatbot counts as transmitting a record “in any manner in which it may be viewed by another person,” which is how the statute reads. One commenter’s version: “Thought crimes are real when you’re sharing your thoughts with Claude.”

* My read between the lines: The legal hinge is whether another person could view the entry, and it only became viewable once Anthropic’s own reviewer opened it. The diary got its reader at the moment of the flag. Whether you like the outcome or not, “damned if you report, damned if you don’t” is now the standing job description for every company running a chatbot, and the privacy policy is the only lock on the diary.

📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — my write-up of a summit where every conversation about AI adoption collapsed into the same word, and that word is the entire story here.

The top one percent of AI payers spend $903 a month, and what they’re buying is tools that get work done, not chat. Viktor is that idea inside Slack: an AI agent connected to 3,000+ of your tools that builds the report, the dashboard, the code or the campaign while you do something else. Not a chatbot. A coworker. New readers get $50 off their first month. Hire Viktor →

ChatGPT Is Forging Cartoonists’ Signatures Nieman Lab

* What happened: Nieman Lab’s Andrew Deck found that ChatGPT, when asked for a “New Yorker-style cartoon,” sometimes signs the result with the real pen name of a working New Yorker cartoonist. A viral Dolly Parton cartoon carried the signature “BLOPER,” which belongs to Brendan Loper, who didn’t draw it. Deck documented more than 15 cartoonists’ signatures appearing without permission or payment.

* Why it matters: A cartoonist’s signature works like a certificate of authenticity, and Loper calls his name “a brand” that “felt very much like a violation of my personhood” when a machine faked it. Emily Flake compared it to being attributed a quote she never said. Condé Nast, The New Yorker’s parent company, signed a licensing deal with OpenAI in 2024, but says it never granted permission to train on its cartoons or use its logo, and the standard cartoonist contract doesn’t allow AI-training licenses.

* What everyone’s saying: OpenAI’s statement says it believes the future of creativity is “fundamentally human” and appreciates the community “flagging bugs and unintended behavior.” After Deck notified the company, ChatGPT began returning a guardrails warning for New Yorker-style prompts, though as of publication it still signs some generic cartoons with real names. OpenAI declined to say how its models might have ingested the signatures.

* My read between the lines: “Bug” is carrying a lot of weight. A model that produces gibberish pen names is guessing; one that produces George Booth’s actual name has seen it. In the interest of disclosure, the cartoons in this brief are AI-drawn and signed with our own website, which is the only signature I’m comfortable letting a model forge.

📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn't Agree To. — the same consent problem, one layer closer to your face: what it means when software uses a version of you that you never approved.

Today’s four bullets are the headline. The paywalled deep-dives are where I take one of these stories and keep going, and members get every one plus the full archive. The daily Brief stays free. <a target="_blank" href="https://artificiallyintimidating.com/subscribe?utm_source=brief&utm_medium=cta&utm_campaign=membership"