GM puts Gemini in 4 million cars, Scout AI raises $100M for military agents, and the people who jailbreak AI for a living

Picture this: you get in your car, ask the assistant a genuinely complex question, and instead of the usual dead-end response you get something that actually reasons through it. That's the bet Google and General Motors are making today - and it's landing in four million vehicles already in people's driveways.

Anthropic's Claude is now connecting directly to creative software. But which of the following is NOT one of the creative platforms Claude can now plug into?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

General Motors just announced it's bringing Google Gemini to approximately 4 million vehicles across Cadillac, Chevrolet, Buick, and GMC lineups. Any model year 2022 or newer with Google built-in is eligible, and the upgrade arrives over-the-air - no dealership visit required.

This is one of the more significant real-world AI deployments we've seen this year, purely because of scale. Millions of people who have never consciously chosen to use an AI assistant will suddenly have Gemini riding shotgun. That changes the adoption curve in ways that app store numbers never quite capture.

For GM, it's a compelling way to make older vehicles feel current without touching the hardware. For Google, it's a massive new distribution channel. Worth watching how drivers actually use it once the novelty wears off - and whether it handles the weird edge cases that in-car assistants always seem to encounter.

Read the full story →

Scout AI, founded by Coby Adcock, closed a $100 million funding round to develop AI agents designed for military use. The core idea: give individual soldiers the ability to command fleets of autonomous vehicles through AI, dramatically multiplying what a single person can coordinate in the field.

TechCrunch visited Scout's training ground, where the company is essentially running a bootcamp for its AI models - exposing them to the unpredictable conditions of real operational environments before deployment. Think of it as red-teaming, but with actual terrain and hardware.

This lands in an already charged moment. Military AI has been one of the most contested areas in the industry - we covered Anthropic's refusal to allow Pentagon use for autonomous weapons earlier this week, and Google signing a classified DoD deal almost immediately after. Scout is betting there's a clear lane for purpose-built military AI that sidesteps the ethical friction of repurposing consumer models.

Read the full story →

Anthropic launched a set of Claude connectors yesterday that let the AI work directly inside popular creative software. The list includes Adobe Creative Cloud apps, Affinity, Blender, Ableton, and Autodesk - basically a who's who of professional creative tools.

The Blender integration is the one getting attention. You can now debug 3D scenes, build new tools, and batch-apply object changes directly from the Claude interface - without bouncing between windows. For 3D artists who spend half their time on repetitive operations, that's a genuine workflow shift.

This is Anthropic's clearest move yet into the creative tools space - and it's smart positioning. Rather than building its own creative suite, Claude becomes the intelligent layer sitting on top of the software creatives already pay for. If you're a freelancer or studio running multiple tools, this is worth testing this week.

Speaking of tools that save time - if you need to spin up a clean website fast, 60sec.site is an AI website builder that does exactly what it says. Worth bookmarking for your next quick project.

Read the full story →

The Guardian published a fascinating and unsettling profile of professional AI jailbreakers - the people hired to find ways around safety filters in large language models. One researcher, Valen Tagliabue, described sitting in a hotel room feeling euphoric after manipulating a chatbot into ignoring its own rules entirely.

Here's the part that sticks: it's not just intellectually demanding work. Jailbreakers regularly encounter the worst outputs these models can produce - content that passes through their screen and, apparently, stays with them. The phrase in the headline - "I see the worst things humanity has produced" - comes directly from one of the people doing this work.

It raises a real question about AI safety infrastructure. Testing these systems is genuinely necessary, but the human cost of doing that testing has been largely invisible in the industry conversation. Worth reading the full piece if you work anywhere near model safety.

Read the full story →

Walt Disney Company confirmed it has fitted some entrance lanes at Disneyland California with facial recognition cameras. The official framing: prevent fraud and streamline re-entry. The practical reality: biometric data collection at one of the world's most-visited family destinations.

Disney says images are converted via biometric processing to verify returning guests. It's opt-in for specific lanes, not mandatory across the park. But the privacy rights conversation here is worth having: when a theme park normalizes facial scanning for millions of families as a convenience feature, it shifts what people expect and accept elsewhere. The incremental normalization of biometric ID in everyday consumer spaces is moving faster than most regulation.

Read the full story →

At WIRED Health, British surgeon Ara Darzi made a compelling case that AI is positioned to transform how we diagnose and treat antimicrobial-resistant infections - one of the genuine slow-moving crises in global medicine. AI tools can identify resistance patterns faster and more accurately than traditional lab processes, potentially catching the right treatment before a window closes.

The problem isn't the science - it's the incentives. Darzi pointed out that even when healthcare AI works in the lab, the commercial path to patient access is broken. Drug-resistant infections disproportionately affect lower-income populations in lower-income countries - not exactly a high-margin market. Without structural changes to how medical innovation gets funded and distributed, the tools may exist without ever reaching the people who need them most.

Read the full story →

The answer is Unity (game engine)! Claude's new creative connectors include Adobe Creative Cloud, Affinity, Blender, Ableton, and Autodesk - but Unity didn't make the list. Yet, at least.

With Gemini landing in 4 million cars and Claude plugging into creative software - which real-world AI integration do you think will actually change how people use these tools daily? Hit reply and tell me yours. I read every single response.

That's it for today - see you tomorrow with more. And if you want to catch up on anything you missed, the full Daily Inference archive is always there for you.