OpenAI released GPT-6 Astra last week, and it’s making me feel two things at once. The capability is real: cleaner code, genuinely polished decks, and a perfect result on a leading cybersecurity benchmark. For an organization running on limited resources, that kind of leap could be an enormous gift.
But after experimenting with it, I keep coming back to a harder question: just because AI can do something, does that mean it should?
These tools are getting good enough that efficiency can now outpace what feels human about the work itself.
Where would you draw the line between a task you’d hand to Astra and one you’d keep firmly in human hands, even if the tool could technically do it?
Wishing you a wonderful week,
Amanda
AI News that Matters
In Plain English
Zero-day vulnerability: a security flaw attackers can exploit before developers even know it exists or have a fix ready.
AI humanizer: a tool that rewrites AI-generated text specifically to slip past AI detectors.
GPT-6 Astra arrives, impressive and a little alarming
OpenAI released GPT-6 Astra, which it's calling its most intelligent and aligned model yet. It completes computer-based tasks in about half the time as its predecessor, writes production-ready code, builds polished documents and decks with real clean designs, and scored a perfect result on a leading cybersecurity exploit benchmark.
Why it matters
I'm watching this one closely instead of celebrating it outright. A model that can find zero-day vulnerabilities on demand is a genuine gift for under-resourced IT teams and a genuine risk in the wrong hands, and OpenAI's own safety notes admit its reasoning is harder to monitor than earlier versions. Impressive doesn't mean uncomplicated. And that same tension applies well beyond cybersecurity: the more capable the tool becomes, the more important it is to decide where human review is non-negotiable.
Source: OpenAI
Try this in 5 minutes
Write one sentence naming a task at your organization you would not hand to an AI model unsupervised right now. Share it with your team this week, even informally.
A few smaller Claude updates worth knowing
Three quieter releases are worth a mention. You can now share a skill or plugin you built with one specific colleague instead of publishing it to your whole organization, so the tool you built for the front desk doesn't have to go to everyone. With the Microsoft 365 connector, Claude can send and read Microsoft Teams messages on your behalf. And work built in Claude Cowork or Claude Code can now be published as an Artifact, a hosted page like a dashboard or reference doc with its own link.
Why it matters
None of these will make a headline on their own, but together they're the kind of workflow detail that decides whether a tool built once actually gets reused.
Source: Anthropic changelog
Gemini Spark can now organize and edit Google Photos
Google's Gemini Spark can now find, enhance, organize, and prepare photos for sharing from a single prompt, while keeping your originals untouched and asking for your confirmation before sharing anything.
Why it matters
A practical, free AI upgrade for anyone using Google Photos. Nonprofits managing event photo libraries or social media content could save significant time organizing and preparing images.
Source: Google
MIT tells schools to stop trusting AI detectors
MIT's Ad Hoc Committee on AI Use released its final report recommending against relying on AI detection tools at all, warning that they trigger an arms race against "AI humanizers" (see the glossary above), produce false positives that land hardest on non-native English speakers and neurodivergent students, and build an adversarial atmosphere of distrust.
Why it matters
Swap "students" for "staff" or "grant applicants" and the same argument holds for any organization tempted to police AI use with a detection tool.
Source: MIT
Free Resources
On-Demand · Traction Rec Webinar Recording
How Recreation Orgs Are Actually Using AI
Traction Rec, the Salesforce-based platform many JCCs and YMCAs already run on, put together a session on how recreation organizations are putting AI to work day to day, not in theory.
Watch the recording
Free Through September · Marketing AI Institute
AI for Marketing Course Series
Five expert-led lessons plus a professional certificate from AI Academy, free to enroll through September 30, with a live Q&A on October 1. If marketing or communications is part of your role, this is worth grabbing before the free window closes.
Enroll for free