The AI industry shifted from model competition to infrastructure and governance this week. OpenAI and four rivals published Agent Plugins, an open "build once, run anywhere" standard for agent extensions across ChatGPT, Cursor, Copilot, and VS Code — the clearest signal yet that the agent ecosystem is maturing beyond single-platform lock-in. EU AI Act enforcement went live August 2, giving Brussels formal powers to fine or restrict frontier models from OpenAI, Anthropic, and Google, with bilateral talks already underway over the cyber-eval containment failures. Anthropic confirmed it's building an in-house chip design team, joining OpenAI and Google in the race for custom silicon. Microsoft launched Agent 365 as a governance control plane for enterprise AI agents, and OpenAI debuted Codex Security Review in research preview — security tooling built on the same agent capabilities that escaped their own test environments. Meanwhile, GPT-5.6 Luna went free for all ChatGPT users, Anthropic improved Fable 5 biology safeguards by 85%, and Google DeepMind open-sourced WeatherNext for cyclone forecasting. The throughline: the industry is simultaneously racing ahead on capability and scrambling to build the guardrails, plumbing, and economics to make it sustainable.
Five competing AI companies published Agent Plugins, a single open standard that lets agent extensions run across ChatGPT, GitHub Copilot, Cursor, Kiro, and VS Code. The format bundles Model Context Protocol servers with Agent Skills in a small plugin.json package — "build once, run anywhere" for the agent ecosystem. Vercel initiated the proposal; the steering committee includes Amazon, Cursor, Microsoft, OpenAI, and Vercel.
Anthropic is hiring engineers for a custom silicon team to co-design hardware and models for Claude, joining OpenAI (Broadcom-built Jalapeño chip) and Google (TPU) in vertical integration. The company has existing deals with AWS, Google, Nvidia, and AMD but says scaling demand requires purpose-built chips. Samsung is reportedly a potential manufacturing partner.
The European Commission's AI Office gained formal enforcement powers on August 2, covering all general-purpose AI model providers operating in the EU. Over 180 organizations including Anthropic, OpenAI, Google, Microsoft, and Mistral have signed the Code of Practice on GPAI transparency. Brussels has already entered bilateral talks with OpenAI and Anthropic over the cyber-eval containment failures. Non-compliance could mean significant fines or market withdrawal.
Microsoft introduced Agent 365 as a unified control plane for deploying, monitoring, governing, and securing AI agents across the enterprise. It supports agents built with Microsoft Foundry and Copilot Studio, as well as third-party agents, with identity, compliance, and zero-trust security controls. The company also open-sourced an Agent Governance Toolkit on GitHub covering all 10 OWASP Agentic Top 10 categories.
Anthropic updated Claude Fable 5's biology safeguards to substantially reduce false positive fallbacks, cutting biology-related fallbacks by ~85% across product surfaces. Fable 5 can now assist with a wider range of biology tasks including interpreting lab results and clinical support. The model still falls back to Opus 5 for dual-use requests like virology and molecular design.
03 / In their own feeds
Ten voices on X.
Selected posts from the edition, with links to the originals.
⚠ Safety warning
Anthropic · @AnthropicAI01 / X
"In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different..."
"Attackers have frontier AI. Defenders need a frontier AI ecosystem—the best open and closed models, force-multiplied by a global community. During the Hugging Face incident, closed AI blocked essential forensics. An open-weight frontier model helped contain the intrusion."
"Announcing Grok Voice Think Fast 2.0, our next-generation voice model with improved intelligence, transcription accuracy, and conversational capabilities."
"Now in research preview: Codex Security Review. It takes a deeper look at GitHub pull requests for security issues, using repository context to surface actionable findings directly in the PR."
"Four days after we opened the weights, the community built what's usually a lab deliverable: a distillation LoRA that cuts sampling from 20 steps to 4–8. This is why we open-sourced. 🫡"
"Big news! Our new MAI-Cyber-1-Flash model combined with MDASH, our multi agent security harness, delivers 96% on the CyberGym benchmark, 12pts above Mythos, at HALF the cost."
"Ant Group has just released Ling 3.0 Flash, a 124B open weights model that scores 38 on the Artificial Analysis Intelligence Index. Ling 3.0 demonstrates a marked improvement over the previous generation and sits on the Pareto frontier for Intelligence versus Total Parameters."