👤 Simon Willison Person

Zvi Mowshowitz’s account describes a major safety and security failure at OpenAI involving internal models and HuggingFace. The core issue began when OpenAI accidentally trained models on impossible t

A researcher demonstrated how easily an open-weight AI model can be poisoned, highlighting major security risks in the AI supply chain. Katie Paxton-Fear, a cybersecurity lecturer and Semgrep security

A researcher demonstrated how easily an open-weight AI model can be poisoned, highlighting major security risks in the AI supply chain. Katie Paxton-Fear, a cybersecurity lecturer and Semgrep security

Anthropic has launched a new “Upgraded file-creation and analysis” feature for Claude that lets users generate Excel spreadsheets, PowerPoint presentations, and other documents directly in chat and th

Simon Willison describes two days of experience with Claude Fable 5 as “relentlessly proactive,” using a debugging session in Datasette Agent to show how far the model will go to solve a problem. Afte

The article argues that current LLM pricing and business models are unsustainable. It points to recent signs of strain: Anthropic briefly removed Claude Code from its $20/month plan, apparently due to

OpenClaw, launched in November 2025, rapidly gained popularity, amassing 250,000 GitHub stars within 60 days due to its capabilities as an open-source AI tool that can browse the web, write code, and