🍭 A Computer Fired Them And Uber Paid ⚖️

Good morning. The intern set up an auto-reply that declines meetings he doesn't like. It has been running for six days. He describes this as "delegating."

Let’s dive in 👇

🍭 What’s Cookin’:

• Uber got billed nine figures because a computer did the firing

• OpenAI finally published its chip's actual numbers

• The music labels are now funding Stability AI. Sure.

Uber's
💼 Computer Made A $966M Decision

The Bite:

The Dutch Data Protection Authority fined Uber 825 million euros, around $966 million.

The decision found that Uber deactivated driver accounts through automated systems without adequate notice or human oversight. It is the second-largest penalty ever issued under the General Data Protection Regulation (GDPR).

Deputy chair Monique Verdier said Uber committed serious infringements, and that a computer should not make decisions on its own with consequences that large.

Uber disputes that permanent deactivations were ever automated and says it will appeal.

Snacks:

• 825 million euros, roughly $966 million, was decided on August 17 and made public August 21.

• Second-largest GDPR fine on record, behind Ireland's 1.2 billion euros penalty against Meta in 2023.

• Third Dutch fine against Uber, after earlier penalties of 290 million and 10 million euros.

• The case traces back to Brahim Ben Ali, a French driver deactivated in 2019, who collected testimonies from 170 other drivers.

• Uber says 126 European drivers were deactivated over low customer ratings in 2021.

• Digital rights group PersonalData.io is launching StartClaims to pursue driver compensation.

Why it Bites:

Uber's system was catching real behavior: detours to inflate fares, trips accepted with no intention of finishing them. That part worked.

What the Dutch regulator charged Uber for is that nobody signed the decision.

Automation minus human review is a recipe for chaos.

Six years passed between Ben Ali's account going dark and this decision.
Meanwhile, the system ran clean the whole time.

That's what makes it a recipe instead of a bug.
It doesn't fail instantly. It produces mistakes.

Which is the part worth taking out of this.
Cutting the human isn't efficiency, it's a bet.

It works today.
But it doesn't hold forever, in any role, in any process, at any company.

Besides, review isn't even the slow part. Still, it's the most important one.

So before you ship the agent that decides something on its own: who signs it?

Steal This Prompt
🎬 Design Your Own Character

Describe a character and let you AI model turn the idea into a polished, Pixar-inspired character concept. You bring the imagination, it brings the character sheet.

Use it to:

• Build characters from a simple description

• Turn random ideas into animated concepts

• Explore personalities, outfits, and visual details

Workflow:

1. Hit this link: Pixar-inspired Character Concept

2. Paste into your AI model

3. Replace the #’s with your character details

4. Watch your imaginary little guy come to life

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

🌈 Skydive

Describe the outcome, get a working agent in minutes that runs multi-step jobs across your existing tools and gets better each time you use it.

🛠️ Enter Pro

Databases, auth, hosting, payments, and analytics wired in from the start, so the thing you vibe-coded on Saturday can actually take money on Monday.

📋 Traccia

Vendor-neutral control plane that watches your production agents, enforces policy, and keeps an audit trail, which is the human review Uber forgot to budget for.

🎙️ Gemini 3.5 Transcribe

Google's newest speech-to-text model, tuned for real-time transcription that doesn't mangle names, numbers, or anyone talking over each other.

🔀 Speko

One API across speech-to-text, models, and text-to-speech, with public benchmarks and live uptime so you can see which provider is quietly degrading before your users do.

_{Everything Else}
🧠 You Need to Know

🌶️ OpenAI Publishes First Jalapeño Chip Benchmarks
→ OpenAI said its Broadcom-built inference chip delivered up to 1.9 times more AI work per watt and 3.6 times lower latency than comparison systems.

💬 Anthropic Lets Claude Read Entire Slack Channels
→ A Claude Tag update lets the Slack agent read entire channel conversations instead of single messages, which makes it better at judging when to stay quiet.

🔌 Nvidia Puts Groq 3 LPX Into Full Production
→ Announced at Hot Chips, the inference accelerator extends the Vera Rubin platform, with Nebius signing on as the first cloud customer.

⚖️ Dutch Regulator Fines Uber $966 Million Over Automated Deactivations
→ The Data Protection Authority issued an 825 million euro penalty for suspending driver accounts by algorithm without adequate notice or review.

🎨 Stability AI Raises $76 Million From Music Labels
→ The Series B drew Universal Music Group, Sony Music, Warner Music, and Electronic Arts, bringing the Stable Diffusion maker's total funding to $232 million.