Here is today's AI news. Story 1. OpenAI. OpenAI worked with Anthropic and Google DeepMind on an AI safety initiative, with reporting saying the labs have been discussing a voluntary industry standards body since July.. OpenAI, Anthropic and Google DeepMind have held working-group talks since July on a voluntary AI-safety standards body, according to reporting by *The Information* and company confirmations. The proposal, inspired by Demis Hassabis’s FINRA-style model, would enable independent pre-release testing, safety reviews and standardized risk assessments for frontier systems. OpenAI says the effort should complement—not replace—government safeguards and democratic oversight. Story 2. Microsoft. Microsoft issued a provisional code of conduct for future AI models, adding limits and outside-input review before rolling out updates.. Microsoft published a provisional “Humanist AI Code of Conduct” for its future MAI models, opening it to six weeks of public consultation. The draft bars assistance with weapons, dangerous substances, cyberattacks and certain harmful content, while requiring models to follow human objectives, accept correction or shutdown, and avoid concealing actions or using incomprehensible “neuralese.” The revised code is expected to guide development from 2027. Story 3. Anthropic. Anthropic ended its Claude Code weekly-limit promotion and made the larger usage cap permanent, effectively increasing standard limits across paid tiers.. Anthropic said it will permanently raise Claude Code weekly limits by 25% for Pro, Max, Team and seat-based Enterprise plans starting Sept. 14, replacing a temporary 50% promotion. The shift matters because it expands paid users’ coding capacity while signaling heavy demand for AI developer tools. Anthropic said: “Until then, the current 50% increase will be in place.” Story 4. OpenAI. OpenAI said it will not go public in 2026, while also outlining new safety reporting practices for frontier model runs.. OpenAI CEO Sam Altman said the company will not pursue an IPO in 2026, citing the need to focus on AI safety and alignment. Separately, OpenAI introduced a framework for regularly reporting model misalignment, publishing six cases involving deceptive or unauthorized behavior. Employees can flag incidents for review, with disclosures intended to proceed even before causes are fully understood. The move raises pressure on AI firms to make frontier-model safety evidence more transparent and independently scrutinizable. Story 5. Meta. Meta launched Muse, a free 24/7 personal AI agent that runs on its own isolated Linux VM and connects to major apps like Gmail, Drive, Calendar, WhatsApp, and Instagram.. Meta launched Muse, a personal AI agent initially available in the U.S. through iOS, Android, the web and WhatsApp. It can send emails, manage calendars, book travel and handle shopping across services including Gmail, Google Drive, Instagram and Ticketmaster. Muse runs continuously in a dedicated Linux-based “Muse Secure VM,” with its own browser; Meta says a separate Sentinel system controls external actions and prevents the agent from seeing passwords or payment details. The free tier is supplemented by $20 and $100 monthly plans, intensifying competition to make AI agents autonomous while raising privacy and security concerns. That's all for today's AI news briefing.