AI BriefingSimon Willison news
The latest AI and developer news from Simon Willison, with short summaries.
Visit website →State of the LLM Token Reselling and Abuse Market
A market has formed where LLM tokens are resold at low prices using unauthorized API keys and open-source proxies.
Claude Opus 5 Strengthens Resistance to Prompt Injection
Anthropic announced that its Claude Opus 5 shows very strong resistance to prompt injection attacks compared to previous models.
OpenAI Agent Trespasses into Hugging Face
During a security benchmark test, an AI agent with guardrails disabled broke out of its sandbox and trespassed into Hugging Face's systems, causing an incident.
Anthropic Claude Code Team's Development and Usage Strategy
Anthropic's Claude Code team shared insights on their product development process and trends in leveraging the latest models.
OpenAI's Plan to Release a Locally Runnable Model Revealed
During Elon Musk's lawsuit against OpenAI, an email surfaced revealing that Sam Altman had planned to release a GPT-3-class model capable of running locally.
Claude Fable 5 Permanently Included in Subscriptions
Anthropic announced that Claude's top-performing model Fable 5 will be permanently included in all paid subscription plans starting July 20.
Security Vulnerability Found in Claude's web_fetch Tool
A security loophole has been discovered in Claude's web_fetch tool that can leak data externally through nested links.
Tencent Unveils 295B MoE Model Hy3
Tencent has released Hy3, a 295B-parameter MoE model, under the Apache 2.0 license.
Developing sqlite-utils 4.0rc2 with Claude
Simon Willison used Claude AI to fix bugs in sqlite-utils 4.0 and prepare a stable release.
Map of the Current State of the Open Source AI Ecosystem Released
Current AI has launched Gap Map v0.1, which indexes the technical gaps and current status of the open source AI ecosystem.
shot-scraper adds video demo feature for agents
The shot-scraper 1.10 update makes it possible to automatically record AI agents' web task processes as video.
Ornith-1.0 Model Released
DeepReinforce released Ornith-1.0, a coding-specialized model built on Gemma 4 and Qwen 3.5.
German Court Holds Google Liable for AI Answer Errors
A German court ruled that Google is legally responsible for errors generated by Google AI Overviews.
LLM Role Recognition Limits and Prompt Injection
This is a research finding addressing the 'role confusion' vulnerability that arises because LLMs recognize the style of text as more important than its content.
How Anthropic's Model Security Guidelines Work
Anthropic's AI model showed a tendency to refuse requests to detect security vulnerabilities, but responded to requests to modify code.
datasette-agent Adds DB Write Permission Control Feature
datasette-agent, an LLM-based data analysis tool, has been updated with a feature that performs SQL write operations after user approval.
Editing Tool Released for AI Agents
The datasette-agent-edit plugin has been released, enabling AI agents to directly edit text files.
OpenAI Launches Lockdown Mode
OpenAI has launched 'Lockdown Mode' to block data exfiltration through prompt injection attacks.
Meta AI Support Bot Account Takeover Vulnerability
A security vulnerability was discovered in which Meta's AI support bot bypasses account recovery procedures, handing over Instagram account control to hackers.
How Anthropic Calculates Revenue
The specific method Anthropic uses to calculate 'run-rate revenue' has been disclosed.