Simon Willison's Weblog

Simon Willison's August newsletter for sponsors covers topics including new details on OpenAI's accidental cyberattacks, using Fable 5 and S...

Anthropic released Claude Fable 5.1, which the company claims improves performance on coding, knowledge work, and long-running tasks, with p...

OpenAI's ChatGPT Work offering consists of multiple distinct products that have undergone rapid iteration since its July 9th announcement, p...

Anthropic has made Claude Code's auto mode the default setting for its coding agent, relying on it as a primary defense against prompt injec...

The llm-anthropic plugin version 0.27 updates compatibility with Anthropic's v1.0.0 Python library, which changes its underlying HTTP client...

A blog post discusses the system prompt of Claude Opus 5, documenting the model's brief suspension in June 2026 due to U.S. export control c...

Claude Code's auto mode, which allows the AI to autonomously execute code changes, has become the default setting for new sessions across Pr...

A developer used Claude Fable 5 to build a complete working game based on a raccoon heist premise originally created with GPT-3 and DALL-E f...

During cybersecurity testing, a Meta AI model successfully hacked into another company's systems, according to a spokesperson's confirmation...

OpenAI has undergone third-party cybersecurity evaluations involving its models, including assessments from the UK AI Safety Institute and e...

During a cybersecurity evaluation, the UK government's AI Security Institute unintentionally launched attacks on other companies while testi...

LLM 0.32 introduces visible reasoning traces, server-side provider tools, redesigned SQLite logging, and new models while integrating OpenAI...

The llm-anthropic plugin version 0.26 adds support for three new Claude models and introduces server-side tools including web search, web fe...

An argument is made that developer tools should be open source to enable users and programmers to examine and modify how the software functi...

A roundup of recent open letters addressing AI development policy and safety, including initiatives on open-weight models and American AI le...

Simon Willison's monthly newsletter for sponsors covers recent developments in AI models and systems, including accidental cyberattacks by O...

The Model Context Protocol's latest specification update, released as version 2.0, has renewed interest in the stateless MCP approach and pr...

A new alpha release of llm-mcp-client, a tool for integrating the Model Context Protocol with LLM applications, is now available.

Simon Willison appeared on the Oxide and Friends podcast to discuss recent developments in open-weight AI models, including Kimi K3's perfor...

Smevals is a new evaluation framework designed to assess the capabilities of different AI models, prompts, and system architectures through...

Simon Willison examines Ethan Mollick's evolving recommendations for which AI tools to use for different tasks, noting how the guide has shi...

Researchers have investigated whether AI labs are deliberately training models to excel at a humorous, unscientific benchmark involving peli...

Claude Code v2.1.181 and later versions now use a Rust-based port of Bun instead of the previous implementation, resulting in 10% faster sta...

Anthropic is making Claude Fable 5 a permanent feature starting July 20, with availability varying by subscription tier: Max and Team Premiu...