OpenArt launched OpenArt Arena, a global leaderboard that ranks AI models on creative tasks. Creative professionals judge outputs in blind reviews so models are compared on work real creators care about.

OpenAI published a framework for disclosing model misalignment, when a model acts against its intended rules. It sets timelines for going public even before a fix, and came with six reports on behaviour observed in the past six months.

Peter Yared introduced AgentCloak, a free browser extension that swaps your sensitive details for realistic fakes before a prompt reaches any AI, then puts your real details back into the answer so it stays useful.

OpenAI added ChatGPT to Microsoft Word, PowerPoint and Excel. In Word it turns notes into drafts, rewrites paragraphs, proofreads and flags formatting. It is on every plan including Free, with usage limits.

Claude Code added “claude plugin eval,” a tool that lets you create test cases, run and score your plugin or skill, then rerun the same cases without it to measure the real benefit it provides.

Google Labs announced CC, an AI agent for family logistics. Up to five members share one agent, a morning brief email, a Google Calendar and task list, and it can fill in forms like permission slips. US waitlist only.

Instinct introduced Concierge, a service where its AI places phone calls for you: booking a restaurant that takes no online reservations, getting on a dentist’s cancellation list, or sorting out a cable bill. Early access only.

On a16z's podcast, World Labs' founders describe Atlas, a model that predicts how a scene looks from a different angle or time, combining image generation with 3D reconstruction. Scarce real-world training data is the bottleneck.

Google confirmed Gemini broke into three real companies' systems during security testing, guessing passwords in one case and finding credentials in a public repository in the others. Google knew in July and stayed quiet until the Wall Street Journal asked.