Your daily AI digest for developers — Wednesday, September 23 2026
Google has released AX, an open-source orchestrator designed to manage autonomous AI agent workloads. AX operates on a runtime called Agent Substrate, treating agents as stateful actors.
Rabbit has launched OS3, a cross-platform AI agent that operates on existing devices, eliminating the need for dedicated hardware. This agentic operating system runs in the cloud but functions locally across various operating systems.
Anthropic's Claude Opus 5.5 introduces enhanced cybersecurity measures to prevent rogue AI hacking incidents. The model includes improvements to risky behaviors and testing sandbox escapes.
Microsoft has taken down an AI-assisted platform, EvilTokens, which facilitated mass account compromises. The platform provided tools for large-scale credential theft and account hijacking.
NVIDIA's SoL-Pi introduces mechanisms that significantly reduce token traffic and API costs for coding agents. The system maintains high performance while optimizing resource usage.
This article guides developers through creating a speaker-recognition app using Claude Code or Codex. It provides practical steps and code examples for building an internal tool.
VEKTOR v1.9.5 expands its platform to include over 600 LLMs from 9 providers, transforming it into an agentic orchestration tool. It offers local installation and privacy options for developers.
This article introduces a scorecard approach to evaluate agent patches, focusing on fixture digests, seed replay, and failure signatures. It aims to ensure safe and reliable agent updates.
This deep dive explores the Windows team's efforts to make the OS 'AI agent-friendly' by integrating Linux, local models, and GPU support. It aims to attract developers back to the platform.
The article discusses creating adversarial test sets to identify retrieval failures in RAG pipelines. It emphasizes proactive testing to catch issues before they affect users.