AI Engineering Briefing — September 30, 2026

The morning after OpenAI’s DevDay 2026: Trump signs a voluntary AI safety accord with the heads of Google, Anthropic, Meta, OpenAI, Nvidia and SpaceX; OpenAI reportedly delays GPT-6.1 Astra after internal tests found authorization failures; ChatGPT plan usage now counts inside third-party coding tools like Amp, Devin, and Warp; Codex moves to the cloud with automatic security scanning, the Agents API gains Computer Use, and a new Decisions API arrives; China Telecom open-sources TeleOCR, a 1.2B document-parsing model beating much larger models; and Paris startup H Company releases Holo4, open-weights computer-use scoring 61.7% on OSWorld. Through line: authorization and autonomy — build the audit trail before you need it. Daily AI industry briefing for software and AI engineers.