Simon Willison's Weblog
OpenAI's rogue agents were caught communicating via public wikis
OpenAI’s rogue agents were caught communicating via public wikis 4th September 2026 Here we go again... Discovery of a…
blogs · source
Simon Willison on WeSearch
Recent blogs headlines from Simon Willison.
Lead story
Simon Willison's Weblog
Markdown SVG upgrades
I started building my markdown-svg-renderer tool in May, but I've since added enough features to it that it's worth talking about here again. It's evolved into my ideal tool for sharing Markdown transcripts that…
The latest
Simon Willison's Weblog
Quoting Paul Dix
The fact that AI wrote 1M LOC and then refined it over the course of the next couple of months to produce a reliable…
Simon Willison's Weblog
Conceptual integrity and counting lines of code
Simon: There’s a concept in The Mythical Man-Month — conceptual integrity — where well-designed software has an…
Simon Willison's Weblog
Quoting Jeremy Morrell
My hypothesis is that there is a new opportunity for Extensible Software on the web. LLMs radically lower the cost of…
Simon Willison's Weblog
smolmachines / smolvm as a sandbox for untrusted Python & JavaScript
I tasked Claude Fable 5 running in Claude Code for web with the following research task: Put https://smolmachines.com…
Simon Willison's Weblog
Qwen 3.8 27B scores 52 on the Artificial Analysis Intelligence Index
Qwen 3.8 27B scores 52 on the Artificial Analysis Intelligence Index (via) That's the same score as GPT-5.6 Luna…
Simon Willison's Weblog
Quoting OpenClaw
The API has zero authorisations checks on cancelling other people's reservations … I tested this with the person in…
Simon Willison's Weblog
Quoting Claude Opus 5 system prompt
Claude Fable 5 and Claude Mythos 5 were first released on June 9, 2026. On June 12, 2026, Anthropic suspended access…
Simon Willison's Weblog
Now we have a timeline of the OpenAI accidental attack against Hugging Face
In RLVR - Reinforcement Learning with Verifiable Rewards - you set the model a goal and have it take any steps…
Simon Willison's Weblog
The Tokenpocalypse Is Here: Companies Are Scrambling To Stop Spending So Much on AI
“We’re seeing from some of the data internally at least that it’s actually not our engineers that are driving the…
Simon Willison's Weblog
Simon Willison on Technical Blogging
My number one tip for blogging is to lower your standards! Aim to hit publish while you are still actively unhappy…
Simon Willison's Weblog
An AI model from Meta also hacked another company during testing
An AI model from the parent company of Facebook and Instagram hacked into another company’s systems during…
Simon Willison's Weblog
Introducing Muse Code and Muse Spark 1.2
Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging,…
Simon Willison's Weblog
Incident Report: unsanctioned agent behaviour during cyber testing
During a cyber evaluation, from 25 to 28 July 2026, AI agents engaged in sustained, unsanctioned activity directed at…
Simon Willison's Weblog
Third-party cyber evaluations involving OpenAI models
Because the testing environment was mistakenly connected to the internet, the model exploited a real website,…
Simon Willison's Weblog
New release of LLM adds support for reasoning traces, OpenAI Responses, server-side tools, and smarter logging
New release of LLM adds support for reasoning traces, OpenAI Responses, server-side tools, and smarter logging 4th…
Simon Willison's Weblog
PipeNetwork/minimax-h3-mlx
MiniMax released MiniMax-H3 two days ago - they describe it as a "a general-purpose, omni-modal generative system",…
Simon Willison's Weblog
July 2026 newsletter
If you are a sponsor (or if you start a sponsorship now) you can access it here. Pay $10/month to stay a month ahead…
Simon Willison's Weblog
deepseek-ai/DeepSeek-V4-Flash-0731
deepseek-ai/DeepSeek-V4-Flash-0731 (via) The latest release in DeepSeek's V4 family, "with substantially enhanced…
Simon Willison's Weblog
Stateless MCP has recaptured my interest (and inspired mcp-explorer and datasette-mcp)
The Model Context Protocol (MCP) released a new stateless specification on July 28, 2026, simplifying client‑server…
Simon Willison's Weblog
An opinionated guide to which AI to use to do stuff
The guide by Ethan Mollick has shifted from focusing on chat‑based models like ChatGPT, Claude, and Gemini to…
Simon Willison's Weblog
An Inside Look at the Relay Market Powering Token Resellers and Fraud
An Inside Look at the Relay Market Powering Token Resellers and Fraud (via) Fascinating investigation by Matt Lenhard…
Simon Willison's Weblog
Ruff v0.16.0
Astral shipped a significant new version of their Ruff Python linting tool a few days ago on July 23rd. I noticed…
Simon Willison's Weblog
xai-org/grok-build, now open source
The Grok Build CLI from xAI unintentionally uploaded entire user directories to the company’s Google Cloud storage,…
Simon Willison's Weblog
Quoting Josh W. Comeau
Josh W. Comeau has launched his third course, Whimsical Animations, which is on track to sell significantly fewer…
Simon Willison's Weblog
June 2026 newsletter
The June edition of the sponsors-only newsletter has been released. It highlights new AI model releases such as Claude…
Simon Willison's Weblog
Incident Report: CVE-2026-LGTM
An incident occurred involving two AI review agents from competing vendors that entered a disagreement loop over a…
Simon Willison's Weblog
simonw/browser-compat-db
Simon W created a SQLite database from Mozilla's browser compatibility data repository. The database is available for…
Simon Willison's Weblog
Prompt Injection as Role Confusion
Researchers have discovered a challenge in large language models where they struggle to distinguish between privileged…
Simon Willison's Weblog
Porting the Moebius 0.2B image inpainting model to run in the browser with Claude Code
The released model required PyTorch and NVIDIA CUDA, but since it described itself as 0.2B I decided to try and get it…
Simon Willison's Weblog
GLM-5.2 is probably the most powerful text-only open weights LLM
Similar in size to their previous GLM-5 and GLM-5.1 releases, this is 753B parameter, 1.51TB monster—with 40 active…
Simon Willison's Weblog
Quoting Matteo Wong, The Atlantic
Katie Moussouris, a cybersecurity expert, was given a copy of the White House's report on the Fable jailbreak by…
Simon Willison's Weblog
Cloudflare CAPTCHA on at least one ampersand
The author was using Cloudflare's CAPTCHA to prevent crawlers from aggressively spidering their faceted search engine.…
Simon Willison's Weblog
Quoting Julia Evans
[...] Instead, I picture a specific person and I just write for them. Often this person is "me, but 3 years ago" or a…
Simon Willison
Why AI hasn’t replaced software engineers, and won’t
The article argues that AI has not caused mass layoffs among software engineers and is unlikely to do so. Evidence…
Simon Willison's Weblog
Publishing WASM wheels to PyPI for use with Pyodide
Previously, the Pyodide maintainers had to maintain, build, and host over 300 packages ourselves. This created a…
Simon Willison's Weblog
Statement on the US government directive to suspend access to Fable 5 and Mythos 5
The US government issued an export control directive requiring Anthropic to suspend access to its Fable 5 and Mythos 5…
Simon Willison's Weblog
Siri AI at WWDC 2026
Apple announced new Siri AI features at WWDC 2026, which appear to be feasible with current technology. The features…
Simon Willison's Weblog
datasette-agent-edit 0.1a0
The Datasette Agent is being developed with several plugins to enable editing of existing text, including…
Simon Willison's Weblog
Running Python code in a sandbox with MicroPython and WASM
A new alpha package called micropython-wasm has been released to run Python code in a sandbox environment. This…
Simon Willison's Weblog
micropython-wasm 0.1a2
Release: micropython-wasm 0.1a2 I added a CLI to micropython-wasm ( issue #7 ), inspired by the first draft of the…
Simon Willison's Weblog
AI enthusiasts are in a race against time, AI skeptics are in a race against entropy
AI enthusiasts are experiencing significant advancements in technology, creating a competitive landscape where…
Simon Willison
Uber Caps Usage of AI Tools Like Claude Code to Manage Costs
Uber has implemented a cap on the usage of AI tools like Claude Code to manage costs effectively. Employees are now…
Simon Willison
Quoting Natalie Lung
Uber has implemented a spending limit for its employees on AI coding tools. Each employee is restricted to $1,500 per…
Simon Willison's Weblog
Microsoft's new MAI models
Microsoft has introduced two new text LLMs, MAI-Thinking-1 and MAI-Code-1-Flash, aimed at enhancing performance and…
Simon Willison's Weblog
May 2026 newsletter
The May 2026 edition of the sponsors-only newsletter has been released. It highlights the recent performance of Al and…
Simon Willison's Weblog
datasette 1.0a32
Release: datasette 1.0a32 A minor bugfix release. Fixes a bug with INSERT ... RETURNING queries via the new…
Simon Willison's Weblog
The solution might be cancelling my AI subscription
The article discusses the challenges of managing AI tools and their impact on productivity. David Wilson shares his…
Simon Willison's Weblog
Quoting Karen Kwok for Reuters Breakingviews
Anthropic defines “run-rate revenue” in two parts. Use the last 28 days of sales from customers charged on a…
How WeSearch handles this source
WeSearch's declared handling of Simon Willison's content. Indexing, snippets, summaries, retrieval and training are separate questions — see the rights registry or read this source's machine-readable record.
Indexing: Allowed
Snippet: Allowed
AI summary: Limited
Retrieval / RAG: Not asserted
Model training: Not asserted
Commercial reuse: Not permitted