MORNING EDITION / 05

The Daily Five.

22 September 2026 · NO

01
Technology
Googlebook pre-orders open, creating a third desktop ecosystem
Google opened Googlebook pre-orders on 21 September, with hardware from Acer, ASUS, Dell, HP and Lenovo. The platform combines Android and ChromeOS technology, runs Photoshop, Lightroom and CapCut, and integrates Gemini into the desktop experience.
Why it matters: This is a serious attempt to make AI part of how the operating system is used, rather than a separate app. It broadens desktop choice, though it is not yet an obvious replacement for a Mac in demanding Resolve work.
Google Blog ↗
02
Software
Neurapix Studio Crop automates consistent portrait crops
Neurapix has released Studio Crop, which generates several predefined crops across an entire portrait series in one operation, such as full-length, half-length and headshot framing. Its AI aims to keep those crops consistent throughout the set.
Why it matters: For volume work such as schools, sports and corporate portraits, this targets repetitive post-production rather than image-making itself. The practical gain is time saved while original files remain editable.
Camera Jabber ↗
03
Gear
Ricoh’s GR IV anniversary edition is a cosmetic limited run
Ricoh Imaging has completed the GR IV 30th Anniversary Edition, limited to 6,000 units worldwide. It uses the same 25.74MP APS-C sensor and 28mm-equivalent f/2.8 lens as the standard GR IV, with cosmetic changes and anniversary accessories.
Why it matters: Buyers are paying for collectability, not better image quality. The edition nevertheless underlines the continuing appeal of a genuinely pocketable fixed-lens camera with physical controls.
PhotoNews ↗
04
AI
OpenAI proposes international testing standards for autonomous AI
OpenAI is urging the US to lead work on international standards for testing increasingly autonomous AI systems, including through AI safety institutes and coordination with China. This is a policy proposal, not a new binding rule.
Why it matters: As models gain access to browsers, terminals, files and purchasing tools, benchmark scores are insufficient evidence of safe behaviour. The key question is how systems perform when they are given a computer and a real-world objective.
Financial Times ↗
05
AI
Meta’s Muse agent pairs automation with a separate control layer
Meta launched Muse earlier this month, and demand in the US now appears to be building. The agent can send emails, fill out forms, book travel and make purchases; sensitive actions are intended to require approval, while the system runs in an isolated virtual machine with a separate control agent.
Why it matters: The notable detail is the architecture: an agent, an isolated computer, an independent safety layer and human approval. That is a more credible model for personal agents than handing one capable model unrestricted access to a machine.
Reuters ↗

Googlebook pre-orders open, creating a third desktop ecosystem

Neurapix Studio Crop automates consistent portrait crops

Ricoh’s GR IV anniversary edition is a cosmetic limited run

OpenAI proposes international testing standards for autonomous AI

Meta’s Muse agent pairs automation with a separate control layer