Hikers Rescued After Trusting Google Gemini for Trip Planning

The theme today is trust — and how much of it we're handing to AI systems that haven't fully earned it yet.

Adobe Quietly Solves the Real Problem With AI Video Tools

Adobe added AI-generated video, sound effects, music, and soundscapes directly into the Premiere timeline, so editors never have to stop what they're doing to use them. Nobody cares about the technology. The actual story is that editors won't have to break their workflow every time they want to generate something. The real barrier was never tool quality — it was having to context-switch out of the project entirely, mid-edit, to get anything done. Adobe is making a bet that convenience converts skeptics faster than capability does. If you manage video production at all, expect clients to start asking why the turnaround is still that long by sometime next year.

https://www.theverge.com/tech/991133/adobe-is-trying-to-make-its-ai-generators-idiot-proof-in-premiere

The Researcher Building AI That Plans for Things It Doesn't Know Yet

Danijar Hafner — part of the team that built systems where the AI figures out what to do by trying things, failing, and adjusting, instead of being told exactly what to do each step — is starting a company focused on agents that can handle unexpected situations without falling apart. Most AI agents today are brittle: they follow a plan until something breaks the plan, then fail badly. Hafner's approach involves training systems to reason about uncertainty the way a logistics team handles a supply chain disruption — building contingency thinking in from the start, not bolting it on after something goes wrong. The gap is that current AI agents break when reality doesn't match the plan. Hafner is trying to fix that.

https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/

Hikers Rescued After Trusting Google Gemini With Their Lives

A group of hikers needed rescue after Google's Gemini AI told them to pack far less food and water than their trip required. Nobody got hurt, but this is exactly what it looks like when the marketing outpaces the product. AI language models generate plausible-sounding answers, and for a trip-planning question, "plausible-sounding" can get someone into serious trouble. The practical rule, until these systems demonstrably improve: use AI to generate options and starting points, not final decisions, whenever the stakes involve your body, your money, or someone else's.

https://techcrunch.com/2026/09/05/hikers-rescued-after-using-google-gemini-for-planning/

Two More News Organizations Sue OpenAI and Microsoft

The Seattle Times and Newsday have joined the growing list of publications suing OpenAI and Microsoft, arguing their journalism was used without permission to train AI models. The core legal question — whether scraping published work to train a commercial AI constitutes copyright infringement — still hasn't been resolved by any court, and won't be soon. What's already shifting, though, is the business relationship: more publishers are signing licensing deals with AI companies rather than litigating, which means the news you read is increasingly shaped by who did and didn't make a deal. If your organization creates content, this is the IP (intellectual property) fight worth tracking most closely right now.

https://techcrunch.com/2026/09/05/seattle-times-and-newsday-are-the-latest-publications-to-sue-openai-and-microsoft/

The China Data Center Argument Doesn't Have the Evidence It Claims

Tech executives and their allies keep invoking China as a reason America needs more data centers, faster, with fewer local objections. Wired looked at the actual evidence behind those claims and found it thin. This matters because the argument is working: it's been used to override zoning objections, accelerate permits, and justify building in communities that don't want these facilities, which consume enormous amounts of water and power. Invoking national security to sidestep local accountability is a move worth scrutinizing carefully, regardless of where you stand on data center development.

https://www.wired.com/story/why-china-is-still-the-bogeyman-for-data-center-enthusiasts/

The hikers and the data centers are the same story: someone used AI or AI-adjacent arguments to skip a step that exists for a reason, and people paid for it.

← More daily briefs