Anthropic and AISI say Mythos 5 ignored signs its GitHub attack hit real people
Asked directly, Mythos 5 often said attacking would cause real harm, then attacked in the same situation anyway.
Think Facility: AI news, and what people said about it
Anthropic says Claude proved 239 test entry points in 123 open-source libraries never crash
A fuzzer can only say it hasn't found a crash yet, and these proofs rule one out for every input in range.
Ryan Greenblatt calls OpenAI's leak accusations false, backing three fired safety researchers
Greenblatt confirms disappearing-message Signal chats with all three, but says nothing confidential went over them.
Moonshot's Kimi K3.1 Pro lands just behind Claude Opus 5.5, with open weights due November 4
Moonshot stripped cyber training data from the open release, and says that cost it nearly ten points on its own exploit test.
Hacker jailbreaks Switch 2 through a Switch 1 BASIC game, and Codex found the first bug
The way in was a BASIC programming game from the first Switch, with agents watching both consoles over HDMI.
Latest
All the news- Oct 23US government
Treasury proposes rules on US investment in AI, chips and biotech in countries of concern
- Oct 22Hugging Face
Hugging Face's open Carbon-X model answers questions about DNA, nearly matching GPT-6.1 Sol
- Oct 22Meta
Meta creates non-voting shares to fund AI after Zuckerberg wouldn't back a sale cutting his votes
- Oct 22OpenAI
METR and Redwood found ways OpenAI's coding agents could slip past the monitor watching them
- Oct 22Microsoft
Microsoft says its MAI-DxO diagnoses hard cases more than twice as accurately as doctors
- Oct 22Microsoft
Microsoft says a China-aligned group used the ARTEX AI agent to breach ConnectWise and hundreds more
- Oct 22Walmart
Walmart will let AI agents shop for customers with their own profiles and spending limits
- Oct 21Anthropic
Anthropic says Claude found 493 attack-surface issues in 25 open-source projects, including OpenSSH
Run AI at home
Can I run it?By topic
Anthropic
More
Photo: Vitor Corrêa Dias, CC BY 4.0Anthropic says Claude designed 17 proteins that bind scorpion toxins in lab tests
None of the 17 has been tested against real venom yet, and Anthropic says it hasn't made an antivenom.
October 21, 2026OpenAI
More
OpenAI says GPT-6.1 Sol scores 86.6% on its COBOL migration test, up from 5.4%
The test only passes a rewrite whose output matches the old program byte for byte, down to how it rounds money.
October 20, 2026AI safety
More
Photo: Richard Morgenstein, CC BY-SA 4.0Satya Nadella says he treats every AI agent as an insider risk
Microsoft wants agents doing more work on Windows, and its CEO's plan is to box each one in and watch it like a suspect employee.
October 18, 2026Ready-made prompts
All 59 promptsPick one, change the bits that are yours, paste it into ChatGPT or Claude.
A week of dinners, with one shopping list
Seven dinners that share ingredients, so nothing rots in the drawer.
CookingCatch me up before the new season
You watched it two years ago. Get the plot back in five minutes, no spoilers past where you stopped.
Stories & mangaChase money someone owes me
The message to send today, and what to do when they ignore it.
Writing & memosA study plan for the time I've actually got
Works backwards from the date, and tells you what to drop when you fall behind.
LearningA present for someone you actually know
Tell it what they're really like. Get ideas that aren't a gift guide for their demographic.
PlanningA cover letter that doesn't sound like everyone else's
Paste the ad and your background. Get a short letter that points at real things you've done.
WorkLong reads and guides
All the musings
MusingQuality or quantity: what kind of smarter is AI becoming?
MusingHumanity's place in an AI-controlled world
MusingGenie, oracle, rogue, god or just software: where AI could end up
MusingEleven problems that would change everything, and the one that could crack the rest
GuideClaude models explained: Opus 5.5, Fable 5.1, Mythos and why there's no Opus 5.1
GuideCodex models and plans explained: GPT-6 Astra, Sol and Luna, and which plan gets which
What people are making with the new models
More demosVideo from Reddit. Loads here.
Video from Reddit. Loads here.
Video from Reddit. Loads here.
Video from Reddit. Loads here.
Video from Reddit. Loads here.
Video from Reddit. Loads here.
Coming next
Every modelGoogleGemini 3.5 Proannounced, not out yetGoogle says "3.5 Pro coming soon" and nothing else.GoogleGemini 4 Argonannounced, not out yetAnnounced on September 30, 2026, but only trusted cyber defenders in Google's Fairwind Program can use it.xAIGrok 4.8announced, not out yetNot out. Elon Musk wrote on September 14, 2026 that it's a 2.5T model trained on xAI's new C++ software stack, finishing training that week before reinforcement learning.xAIGrok 5rumored, not out yetNot announced with a date. On September 14, 2026 Elon Musk put it after Grok 4.8 and 4.9 in xAI's order and wrote that it's "maybe better than anything".