The Daily AI Drop - Morning Edition, November 12, 2026

A Google-flavored morning: big security claims for Gemini 4 Carbon, a fast and cheap new model put through its paces, a challenge to redesign work rather than just automate it, and a real-world Codex story from the travel industry.

Listen: Listen to this episode

Google says its AI is immune to prompt injection

Channel: aish_reganti

Video: Watch source video ↗

Google claims Gemini 4 Carbon blocked every attack on the Gray Swan prompt injection benchmark — a 0% attack success rate. If that holds in the real world, it's a genuine milestone, since prompt injection has been AI security's open wound for years. But a benchmark pass doesn't make every app built on the model secure: EchoLeak, the first known zero-click prompt injection exploit in a production AI system, got a single ordinary-looking email past Microsoft 365 Copilot's injection classifier and made it leak data through an auto-loaded image. Benchmarks measure the model; attackers target the app.

Google's Gemini Decision 1: Fast, Cheap, Not Perfect

Channel: fahdmirza

Video: Watch source video ↗

Fahd Mirza tests Google's new lightweight Gemini Decision 1 model across image, text, audio, video, and PDF tasks, plus benchmarks. It doesn't try to beat the flagships — the question is whether it's good enough for everyday jobs where speed and cost matter more than squeezing out the last drop of quality. For summarizing documents, triaging images, and routine audio, a fast, cheap, good-enough model changes the economics of the whole product.

Let AI redesign your work

Channel: matthew_berman

Video: Watch source video ↗

Matthew Berman argues most people use AI to do their old job a little faster, when the real opportunity is to rethink the job itself. Instead of asking how AI can help write the same report in half the time, ask whether the report should exist at all — or whether the whole workflow could look completely different. Automating a bad process just gives you a faster bad process; the bigger wins come from redesigning work from scratch with AI as a collaborator.

What Codex Unlocks for Holiday Extras

Channel: OpenAI

Video: Watch source video ↗

Holiday Extras, the travel company, used Codex to rebuild its main app in two weeks — and the story isn't just about speed. Codex is helping the team respond to customer feedback faster, and the people who understand customers best are now the ones building the software to serve them. CEO Matthew Pack, David Lee, and Liam Thompson talk about giving colleagues the confidence to build. For a travel business where the experience lives or dies on little details, that combination of speed and closeness to the customer is exactly where AI coding tools pay off most visibly.