🍭 It Fought Back With China's AI 🧠

Good morning. Hugging Face got broken into by an AI agent that never checked in with a boss, never slept, and never once wondered if it should.

Let’s dive in 👇

🍭 What’s Cookin’:

• Hugging Face got hacked by an AI agent working entirely alone

• Google just got ordered to open Android to rival AI assistants

• Alibaba dropped a model it says only Fable 5 beats

Cyber Security
🚨 Hacked By An AI Agent Working Alone

The Bite:

Hugging Face disclosed on July 16 that its production infrastructure was breached by an autonomous AI agent system, with no human operator involved end to end.

The attacker exploited two code-execution flaws in Hugging Face's dataset processing pipeline. One was a remote-code dataset loader, the other a template-injection bug in dataset configuration.

From that foothold, the agent escalated privileges and harvested cloud and cluster credentials. It then moved laterally across several internal clusters over a single weekend.

When Hugging Face's security team tried commercial frontier model APIs, including Anthropic's and OpenAI's, to analyze the attack, safety guardrails blocked requests containing the real exploit payloads and command-and-control data.

Snacks:

• The attacker's agent ran more than 17,000 actions across short-lived sandboxes.

• Hugging Face found no evidence public models, datasets, Spaces, or its software supply chain were altered.

• The company pivoted to GLM 5.2, an open-weight model from China's Z.ai, running it on its own infrastructure.

• Hugging Face reported the incident to law enforcement and hired outside forensic specialists.

• The company is urging users to rotate access tokens and review recent account activity.

Why it Bites:

The attacker's agent didn't need a plan approved, a policy checked, or a reason.
It just kept running for a weekend.

Hugging Face's own defenders didn't have that freedom.

When they asked frontier commercial models to help read the attack logs, the safety filters blocked them. The same content that makes an exploit an exploit made it look like an attack instead of an investigation.

So the company that got hacked ended up defending itself with a model built somewhere the guardrails don't reach.

That's the actual risk here, and it has nothing to do with dataset loaders.

Every security team now has a reason to keep an unrestricted model on standby for the day their own vendor won't help them.

And the next incident won't wait for anyone's filters to catch up.

Steal This Prompt
🎭 Shadow Puppet Theatre

Turn any idea into a dramatic shadow-puppet performance that looks like it escaped from a centuries-old theater.

The prompt is built for cinematic image and video generation with a theatrical silhouette aesthetic.

Use it to:

• Reimagine movie scenes as shadow puppet performances

• Turn fantasy creatures into eerie silhouette art

• Create thumb-stopping social posts with a timeless aesthetic

Workflow:

1. Hit this link: Shadow Puppet Theatre

2. Paste into your AI model

3. Replace the #s with your scene, characters, or story

4. Watch it stage a puppet show your art teacher would've framed

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

🔥 Fuzzy AI

Finds people already talking about your market, warms them up with comments and content, then fires off LinkedIn and email sequences that sound like you wrote them.

🗂️ Nautis

Runs fundraising, finance, hiring, and CRM off one shared AI brain instead of a founder's fifteen disconnected SaaS tabs.

🐛 Replay QA

Explores your app like a real user, records every session, and hands your coding agent the exact bug and the fix before your customers find it first.

🎙️ Skippr AI

Embeds a live AI agent in your product that sees the screen, talks in 10 languages, and onboards users on its own. Two lines of code and it's live.

🎨 Loova AI

Spits out UGC ads, product commercials, and avatar videos for under $2 each, built to convert rather than just look pretty.

_{Everything Else}
🧠 You Need to Know

🏛️ EU Forces Google To Open Android To Rival AI
→ The European Commission ordered Google to give rival AI assistants the same Android system access as Gemini and share anonymized search data with competitors starting January 2027.

🚨 Autonomous AI Agent Breaches Hugging Face Infrastructure
→ An AI agent system with no human operator exploited two code-execution flaws in Hugging Face's data pipeline, harvesting cloud and cluster credentials over a single weekend.

🧠 Alibaba Claims Qwen3.8 Rivals Claude Fable 5
→ Alibaba's 2.4-trillion-parameter Qwen3.8-Max preview launched July 19 inside paid products, with the company claiming performance second only to Claude Fable 5 ahead of a planned open-weight release.

🌐 Current AI Raises $400M For Public AI Infrastructure
→ The nonprofit secured $400 million from governments, foundations, and tech companies to build open AI infrastructure for underserved languages, including an offline AI device supporting 22 Indian languages.

💾 SK Chief Warns AI Memory Shortage Could Turn Geopolitical
→ SK Group Chairman Chey Tae-won said customers are asking for 60 to 100 percent more AI memory in 2027, warning of coming government-to-government pressure over supply.