KORDUROYTuesday, September 8, 2026 · Run 2 · 4:00 PM ET
The eco-scam app, Astra's opacity problem, and writing in your own voice
3 stories from 1 source
TL;DR
- A viral "eco-friendly" AI app got tagged a scam for reviving a debunked water-scarcity myth.
- GPT-6 Astra's new "opaque recurrence" training method has safety researchers alarmed about lost transparency.
- ChatGPT Work can now learn and reproduce your writing voice across Gmail, Slack, and Drive.
AI Tips1
Make ChatGPT Work write in your voice
- Open ChatGPT, toggle on Work, then go to Settings > Personalization > Writing Style.
- Connect the apps with the best examples of your writing, such as Gmail, Google Drive, and Slack.
- Let ChatGPT analyze the patterns in your writing, then return to Work and ask it to draft an email, message, or document as usual.
- Review the output to confirm it sounds like you, and adjust the connected sources if it doesn't.
Source: Superhuman AI
AI News2
Viral "eco-friendly" AI app draws scam accusations
EcoGPT has been downloaded more than 100,000 times on the back of videos claiming AI data centers will make fresh drinking water disappear, a claim Superhuman AI says has been "repeatedly debunked." Critics are now calling the app a scam that uses greenwashing to exploit environmental guilt.
Why it mattersExpect this specific water-scarcity myth to keep resurfacing in client conversations about AI's environmental cost; worth having the debunk ready rather than letting it pass unchallenged.
Source: Superhuman AI
GPT-6 Astra's efficiency trick worries AI safety researchers
Reporting ties Astra to a training method called "recurrent depth" (or "opaque recurrence"), which cycles a query through the same internal layers repeatedly instead of reasoning in visible steps. Redwood Research's chief scientist Ryan Greenblatt called it "the single worst development for AI security/safety to date," though OpenAI says Astra's chain of thought "remains largely readable" for now.
Why it mattersChain-of-thought transparency is a selling point vendors use in AI-governance pitches; this is an early sign providers may trade it away for performance, worth flagging in any client AI-risk discussion.
Source: Superhuman AI
Summarized from Superhuman AI (Sep 8)
