AI News Daily Digest (26-09-09)

Google’s AlphaGenome Atlas aims to predict every possible DNA letter change

Google DeepMind unveiled AlphaGenome Atlas – a “predictive map” intended to forecast how any single DNA change could affect biology. The goal is to turn genome variation into something usable for researchers, potentially accelerating the path from molecular insight to new treatments.

Read the full article here

Adobe is trying to make its AI generators “idiot-proof” inside Premiere

Adobe’s Generative Media for Premiere focuses on keeping editing inside the timeline by generating context-aware clips directly where editors need content. Instead of bouncing between apps for assets, users can highlight a gap and generate editable video, audio, and soundscapes without breaking workflow.

Read the full article here

OpenAI expands “supporting journalism” from classrooms to newsrooms

OpenAI is rolling out tools, training, and partnerships aimed at students, educators, journalists, and newsroom organizations. The push frames journalism support as both capacity-building and practical adoption – helping reporters and institutions use AI more effectively and safely.

Read the full article here

ChatGPT Images 2.5 brings more personalization from sketches and reference photos

OpenAI’s ChatGPT Images 2.5 targets higher-quality, more tailored results by letting users combine ideas, rough sketches, and reference photos. The update is positioned as a step toward images that better track what you meant – not just what you described.

Read the full article here

OpenAI’s Navier-Stokes work shares an approach to the Millennium Prize Problem

OpenAI published a writeup describing an AI-generated solution attempt for the Navier-Stokes Millennium Prize Problem, including a formal proof in Lean. The publication is aimed at turning a long-standing theoretical challenge into something verifiable with modern proof tooling.

Read the full article here

Anthropic faces a class action over subscription limits for power users

A group of Claude subscribers is suing Anthropic, alleging the company misrepresented how much value customers would get from its top-tier Max subscription. The lawsuit claims users were effectively cut off from popular applications while being sold a tier advertised as providing more extensive access.

Read the full article here

Safety for Whom – refusing the “right subset” instead of shutting the whole topic

A Hugging Face post argues for a more surgical approach to AI safety – refusing the specific problematic subset of a topic while allowing the rest. The idea is to reduce overblocking and make refusals align more precisely with real risk boundaries.

Read the full article here

This AI entrepreneur is building agents that plan ahead for the unexpected

Danijar Hafner’s new effort focuses on AI agents that can anticipate and adjust when reality diverges from the plan. The reporting ties the vision to the broader push for more capable agents – ones that don’t just execute tasks, but manage uncertainty.

Read the full article here