Claude desktop app now supports third-party API proxies
Anthropic has quietly added official support for third-party API proxies in the Claude desktop app. Here's how to set it up.
Developer
Most Recent
AgentFlow AI System Finds Critical Chrome Zero-Days, Including Sandbox Escapes
An AI system called AgentFlow automatically coordinates teams of AI agents to find software bugs, discovering 10 zero-day vulnerabilities in Chrome—including two critical sandbox escapes—in a week-long test.
Perplexity's open-source search agent beats GPT-5.4 on accuracy and cost
Perplexity details its two-stage training process for web search agents, showing its open-source Qwen-based model outperforms GPT-5.4 and Claude Sonnet on accuracy while costing a fraction of the price.
Vercel's security breach was bigger than it first looked
Vercel's investigation into a recent security breach found the attack was more widespread than initially reported, involving malware distribution and uncovering older, unrelated account compromises.
OpenAI's Codex team fixes a sneaky bug that was making its agents lazy
A silent authentication bug in OpenClaw was causing OpenAI's Codex agents to fall back to a less capable runtime, making them lazy. A simple fix unlocked their full potential.
OpenClaw's rapid updates are breaking things for Nvidia, ByteDance, and Tencent
Major tech firms are stuck on an old version of OpenClaw, fearing that its rapid-fire updates will break their products. The project's maintainers are divided over whether to adopt more corporate processes.
Xiaomi's MiMo V2.5 AI Model Goes Public, Claims to Rival Claude and GPT
Xiaomi's new MiMo V2.5 AI models are now in public testing, with the top model targeting Claude Opus and GPT-5.4 performance. It aced complex coding tasks and will soon be open-sourced globally.
Hugging Face's new AI agent reads papers, picks data, and runs your ML experiments
Hugging Face open-sourced ml-intern, an ML research agent that autonomously reads papers, curates datasets, launches GPU training, evaluates results, and iterates — all without human intervention.
OpenClaw founder: Anthropic said yes, then blocked us anyway
OpenClaw founder Peter Steinberger confirms Anthropic's Claude Code lead gave verbal approval for CLI use, but server-side fingerprinting still blocks the tool.
OpenClaw's latest update patches a security hole and switches to GPT-Image-2
OpenClaw's new version integrates OpenAI's GPT-Image-2 for image generation by default and fixes a security vulnerability that could have allowed unauthorized command execution.