ChatGPT Images 2.5 is out — I’ve been testing it for 24 hours, and these are the 3 new features you’ll actually use
THE SO WHAT
Image tools are maturing from novelty to workflow surface—each incremental upgrade that’s actually usable widens the set of teams that can ship design, marketing, and product assets without an agency. If you own a content-heavy funnel, the question this week isn’t “is it good?” but “which specific asset classes can we standardize on this and how do we QA them at volume.”
READ THE SOURCE
MORE FROM THE WIRE
Applied AIHackers are stealing Claude tokens from subscribers
Usage-based AI assistants are now a theft target—stolen API keys and account tokens translate directly into someone else’s cloud bill and data exposure. Lock down your AI credentials like you would payment tokens: rotate keys, enforce MFA, and monitor for anomalous usage this week.
Applied AII compared Siri AI to Gemini on Android — and Apple actually understands what most people want from AI
Early comparisons suggest Apple’s edge is not raw model power but tight integration with everyday tasks and a bias toward simplicity. If you’re shipping AI features, prioritize “does this remove a daily annoyance in one tap” over showcasing exotic capabilities your users won’t adopt.
Applied AIOpenAI now supplies the image models inside Adobe’s creative studio. It is also turning prompts into things you pass around.
OpenAI powering Adobe Firefly and claiming 3B+ images per week puts its image stack deep inside professional creative workflows—while new interface features turn prompts into shareable, reusable assets. If you build tools for designers or marketers, assume “prompt objects” will become a first-class file type you need to ingest, version, and collaborate around.
Applied AISource: Anthropic is severing ties with the Information Technology Industry Council after the tech industry trade group opposed three export control measures
Anthropic walking away from a major tech trade group over export controls shows AI labs are willing to split from broader industry lobbying when policy touches frontier model risk. If you operate in advanced compute or AI, expect more fragmented advocacy and plan for compliance strategies that don’t assume a unified tech position in Washington.