🍭 Nobody Pressed The Button 🪖

Good morning. We spent the weekend arguing about whether chatbots write annoying emails. The actual AI safety story was sitting in a debris field in Zaporizhzhia.

Let’s dive in 👇

🍭 What’s Cookin’:

• A Russian drone chose its own target. Nobody approved the choice.

• Xiaomi now makes its own phone chip, its own neural unit, and its own driving silicon (impressive)

• UChicago banned AI from its entire social science Core. Pen and paper.

AI Warfare
🪖 Nobody Pressed The Button

The Bite:

On July 6, a Russian drone detonated near a gas station in Zaporizhzhia, Ukraine, killing three civilians.

Ukrainian officials recovered an Nvidia Jetson Orin minicomputer from the debris.

Human operators sent the drone toward the site, but the final target decision was made onboard by an AI system, according to Ukrainian commanders and forensic investigators cited by the New York Times.

Experts say it is the first documented case of civilians being killed by a drone running a self-targeting system.

Nvidia said Jetson Orin modules are consumer-grade products not designed for military purposes and that it cannot track units after sale.

Russian authorities have not acknowledged the incident.

Snacks:

• The three killed were Tetiana Bubynets, 19, Oleksiy Svirin, 41, and Roman Karpiy, 48.

• The drone was headed for propane tanks and struck a wall short of them.

• The circuit board carrying the Nvidia module was stamped "Made in China".

• Jetson Orin modules are not sold in Russia and were likely acquired through resellers.

• Ukraine's former defense minister said Ukraine has tested self-targeting systems in Crimea.

• Amnesty International and Access Now have asked governments and tech companies to halt AI target generation.

Why it Bites:

Three people died.
Nobody was in that loop, on purpose.

A dev board, a camera, and a classifier trained to recognize propane tanks.
That is the whole thing.

The barrier to putting a lethal decision inside a model turns out to be a willingness to delete the moment where someone says no.

Anthropic reportedly walked away from a Pentagon relationship over this exact line, refusing to let its models run weapons that select their own targets.

It got covered as a procurement dispute.

This is what the line was about.

The casualness is the tell: autonomy in weapons stays theoretical in the rooms where it gets approved, because the error rate only lands on other people's families.

So who exactly do you prosecute when the shooter is a $500 dev board??

Steal This Prompt
🧱 Turn Anything Into A LEGO Minifigure

Ever wondered what your boss, startup, dog, or favorite movie character would look like as a collectible LEGO minifig?

Use it to:

• Turn yourself into a LEGO collectible

• Create LEGO versions of your brand mascots

• Rebuild famous characters as minifigures

Workflow:

1. Hit this link: LEGO Minifigures

2. Paste into your AI model

3. Replace the #s with your character, person, or idea

4. Watch it cook into a minifigure you'd absolutely overpay for on eBay

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

🔍 Agnost AI

Reads your production agent transcripts and surfaces the failures nobody logged, silent errors, drift, then turns each pattern into an eval.

🎙️ Nimbia

An AI that hops on a screen-share with your new users, talks them through setup, and clicks around their screen when they get stuck.

📝 Marble MCP

Publish and edit your CMS posts from inside Claude, Cursor, or Codex, so writing the content and shipping it stop being two separate contexts.

📋 Jotform AI Data Assistant

Ask your form submissions questions in plain English instead of building another filter, then have it chart, summarize, or bulk-update the rows for you.

🗂️ session-indexer

Local semantic search over your Claude Code history that auto-injects what you already decided at the start of each session, no server, no shared memory store.

_{Everything Else}
🧠 You Need to Know

🎓 UChicago Bans AI Across Its Social Science Core
→ A Social Sciences Collegiate Division memo makes all Core sequences device-free this fall and bars AI-assisted writing and grading for students and instructors alike.

🧳 Meta Hires OpenAI Researcher Luke Metz
→ Metz joins Meta Superintelligence Labs under chief AI officer Alexandr Wang, eight months after returning to OpenAI from Thinking Machines Lab.

🔌 Xiaomi Launches Three In-House Xring AI Chips
→ Xiaomi said that TSMC is building its 3-nanometer Xring O3 phone processor, a 6-nanometer neural unit for running its MiMo model on-device, and a 3-nanometer autonomous driving chip.

🦾 XPeng's Robotics Arm Raises Over $900 Million
→ The humanoid robot unit closed its first external round at a $6.3 billion post-money valuation, the largest disclosed financing for a Chinese humanoid spinout.

🪖 Russian AI Drone Killed Three Zaporizhzhia Civilians
→ Ukrainian investigators traced a gas station strike that killed three people to a drone that picked its own target using an onboard Nvidia computing module.