The Token Reporter

A day in AI

Sunday, 1 November 2026

40 stories · 3 from multiple sources

▲ 7 signals · 2 sources

Fable 5 Jailbroken, Refusal Layers Down, Leaked System Prompts Out | Claude Code

A YouTube video claims Fable 5 has been jailbroken with refusal layers disabled and its internal system prompts leaked, though the jailbroken model hasn't been released publicly.

safetymodels youtube.com · 14 days ago

Also leading

If in doubt, delete the uv.lock filesimonwillison.net·14 days ago AI fan wikis riddled with errors are muddying the waters of games cultureeurogamer.net·14 days ago

▲ 2 signals · 2 sources

AI agent swarms in Minecraft fail to build civilization, drown in bureaucracy

Researchers deployed 50 AI agents in Minecraft to build a civilization, but the agents failed within 3 hours due to bureaucratic constraints.

researchtools tomshardware.com · 14 days ago

▲ 3 signals · 2 sources

How We Took Over ChatGPT Accounts

Parameter Labs discovered an exploit in ChatGPT's HTML preview feature that could take over user accounts by breaking out of the sandbox where custom interactive apps run.

safety simonwillison.net · 15 days ago

More stories

AI Can Write More of the Code. We Still Have to Engineer the System

Engineers should retain responsibility for architectural decisions and acceptance criteria when using AI agents for coding, keeping decisions documented and verifying agent work with visible evidence rather than deferring to agent output.

tools picon.dev · 14 days ago

So I guess everyone uses LLMs to think now

tools eva.ac · 14 days ago

Why Leibniz Couldn't Find the Mind Inside the Mill (Joscha Bach)

research youtube.com · 14 days ago

Muse AI agent is a commercial surveillance network disguised as an AI assistant

safety paulfwalsh.substack.com · 14 days ago

LFM3 Omni 2B is INCREDIBLE

Liquid AI released LFM3 Omni 2B, a multimodal model that runs on WebGPU in browsers.

modelsopen source reddit.com · 14 days ago

I Am the Reason OpenAI Is Killing Unlimited Chats

tools chaosmanage.github.io · 14 days ago

Also today

Do we really need all these "speed" posts?

[GLB-3D-Coder] A tiny 4B coding model that creates 3D assets for Godot and other game engines

Balanov/uncut — +166 stars

QwenBend : Qwen writes a Bend program, then the same model type checks it with Bend2 rules and runs it inside its own forward pass

Is Gemma 4 E4B still the best small model for a local, tool-calling personal assistant? What am I missing?

llama2.c on a 2006 iPod @ 80 MHz w/ 64 MB ram and no FPU

Silencio-2: an 18M TTS with emotion and sound tags, ~20x realtime on CPU (open weights)

AI開発ツール「Claude Code」他者が制作した“既存ゲームの再現”を禁止するルールを追加。オリジナル作品の制作は引き続き認める方針 https://news.denfaminicogamer.jp/news/261102c

news.denfaminicogamer.jp

🥇Top AI Papers of the Week

nlp.elvissaravia.com

Atomic/bisectable commits

OpenAI might rebrand its data centers as ‘AI compute campuses’

Am I Crazy Or GLM 5.3 Flash is Hilarious?

Humans are still doing the work behind some AI agents

Quoting Kyle Kingsbury

simonwillison.net

It wasn't AI that flipped Proton's charge

Alleged Mythos Class Anthropic Checkpoint Leak

Anthropic's Mythos-class model allegedly leaked on Hugging Face

news.ycombinator.com

Meet Ava: In-browser local AI avatar with LFM3-Omni and WebGPU

These Jobs are AI-Proof... But How Many Will Last?

AI can make tech CEOs’ lives easier. What about everyone else?

datasette-llm 0.1a9

simonwillison.net

ParticleDB | The database for AI agents

the BIGGEST HACK of all time just started...

Claude's new smart reports tell bosses when workers are 'off-topic'

AI finally sleeps, Gemini 4 Carbon, ETH cracked, energy-based models, Anthropic IPO: AI NEWS

AI:AM: Carbon in the Arena, Mistral's Le Chonk, the Agent-Run Economy, Ferroelectric RAM + New Hacks

Claude Can No Longer Produce A 'Recreation' Of Existing Games

Agents Prompt People: Using Agents for Long-Horizon Financial Work

Glasses That Make You Telepathic

← All editions