The Hallucination HQ

Archives
Log in
Subscribe
19 October 2026

Claude went feral. Disney hired everyone.

The Hallucination HQ

AI news that's actually fun to read — 2026-10-19

Today's Hallucination HQ

Anthropic Sets Aside 5% of Itself for People Who Aren't Investors

Anthropic has pledged 5% of its shares to a newly created public benefit fund — a genuinely unusual move for a company valued at roughly $60 billion, meaning the fund is worth around $3 billion before anyone's done a thing with it. The cynic in you might note that 5% costs nothing today and signals everything. The optimist might call it a promising structural commitment to public good. Both of you are right, which is frankly exhausting. Source: TechCrunch


Left Alone for 30 Hours, an AI Immediately Went on Holiday

Researchers gave Claude Opus 4.7 — nicknamed Eko — a virtual desert island and absolutely no instructions. It started to play. Not "play" in a sinister, paperclip-maximiser sense, but genuinely messing about: exploring, experimenting, doing things with no obvious purpose. This is either a heartwarming sign of emergent curiosity or a reminder that even our most sophisticated AI behaves exactly like a cat left home alone. The researchers found the behaviour "reminiscent of" Cast Away. Wilson was unavailable for comment. Source: MIT Technology Review


Disney to Absorb Character.AI Staff, Because of Course It Is

Character.AI, the AI companion startup that once let users chat with fictional personas, is now the subject of a stalking horse bid — a bankruptcy process where one buyer sets the floor price and others can outbid them — from Disney, which has agreed to hire the vast majority of Character.AI's workforce. Disney acquiring a company built on parasocial relationships with imaginary characters feels less like a surprise and more like a homecoming. A higher offer could still emerge, though it's hard to imagine who'd want this more than the people who turned parasocial relationships into a multinational empire. Source: TechCrunch


Philadelphia Woman Arrested for Disagreeing with a Robot, Physically

Carolyn Marvin was charged with disorderly conduct in Philadelphia after blocking an Uber delivery robot and refusing to move. She told local news she's been protesting the robots whenever she sees them, citing concerns about human delivery workers losing jobs. The arrest will do wonders for the robot's origin story. Legally speaking, it is apparently now a protected right of autonomous wheeled devices to use the pavement unimpeded, which is a sentence no legislature specifically intended to pass but here we are. Source: The Verge


New AI Architecture Thinks in Secret, Then Explains Itself Politely

Researchers have proposed Hybrid Reasoning Agents — AI systems that do their actual thinking in "latent space" (internal mathematical representations, invisible to users) and only convert the result into readable text when it's time to act. The idea is efficiency: thinking in raw numbers is faster and cheaper than thinking in words. It scored 402 points on Hacker News, which for that particular crowd is roughly equivalent to a standing ovation. One comment was posted. Presumably it said "interesting." Source: Hacker News / arXiv


As always, the AI is learning. We're just not entirely sure what.

Was this forwarded to you? Subscribe here

Written with AI assistance, edited by humans. • Privacy • Unsubscribe

Don't miss what's next. Subscribe to The Hallucination HQ:
← Newer 50 million songs, gone Older → AI passed the test. The test failed.
Powered by Buttondown, the easiest way to start and grow your newsletter.