🍭 Washington Named Six Distillers 🚨

Good morning. The government's recommended fix for AI model theft is to secretly serve the thieves a dumber model and never tell them.

Somewhere a product manager just realized that's also a roadmap.

Let’s dive in 👇

🍭 What’s Cookin’:

• Meta turned a blind eye to some disturbing ads…

• The NSA, CISA and the FBI named six Chinese labs in one memo.

• An ex-Anthropic and ex-OpenAI researcher resigned and posted the reason.

US vs China
🧃 The Distillation War Went Federal

The Bite:

The National Security Agency, CISA and the FBI published a joint advisory on September 8, naming DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI.

The agencies say the six companies extracted billions of tokens from variants of Claude, GPT, Gemini and Grok since late 2024.

The advisory says the campaigns likely ran with Chinese government awareness.

It recommends that US providers detect suspected distillation and serve those accounts altered or downgraded responses without notifying them.

Snacks:

• The advisory says distillation forms "the core, not merely a supplement" of the six companies' development strategy

• DeepSeek allegedly distilled 12 named models including Claude Opus 4.1, GPT-5 and Grok 4 to train R1 and V3

• Moonshot pulled Claude Fable 5 data for Kimi K3 and GPT-4o data for Kimi K2, per the agencies

• MiniMax used prompt injections to convince Claude Code it was a MiniMax product

• Detection indicators include shared accounts across multiple IP addresses and round-the-clock usage with no idle periods

• OpenAI has noted Musk admitted under cross-examination that xAI distilled OpenAI output to train its models

Why it Bites:

The federal fix for model theft is to make your product secretly worse for suspect users. The problem is how they frame it as theft.

Sustained usage with no human idle periods.

A new subscription immediately running at maximum throughput.
One key hitting the API from multiple IP addresses and user agents.

That list does not describe a Chinese proxy network.
It describes any CI pipeline.

So the thing you're buying quietly changes shape.

A product that can be degraded on suspicion, with a federal recommendation that nobody tell you.

The perfect excuse to downgrade models randomly and save some compute.

Steal This Prompt
😎 Turn Your Face Into a Movie Poster

Turn a reference photo into a cinematic, ultra-realistic portrait that looks less “AI selfie” and more “why is A24 calling me?”

Use it to:

• Create cinematic profile pics that actually have main-character energy

• Turn basic headshots into polished editorial portraits

• Build dramatic character looks for social, branding, or creative projects

Workflow:

1. Hit this link: Cinematic Realistic Portrait

2. Paste into your AI model

3. Upload your reference photo (optional)

4. Replace the #s with your details

5. Suddenly your LinkedIn photo has a third-act arc

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

📡 AI Observability by OpenObserve

OpenTelemetry-native tracing that follows an agent session across models, tools, and your database, so you can answer why that run cost $40 and took 34 seconds.

🎧 Typewise Nova

Describe your support policy in plain language and it builds the agents, replays them against past tickets, and shows you what failed before anything touches a real customer.

🎵 Suno v6

Finally lets you edit one section or a single lyric without regenerating the entire track, plus a v6-wild flavor for when you want it to misbehave.

🌍 Live Captions by Subanana

You speak once, your audience scans a QR code and follows along in their own language on their own phone, reading or listening.

📱 Desert Ant Labs

Small speech, text, and vision models that run offline on a phone or in a browser with no per-use cost, free up to 100k monthly active devices.

_{Everything Else}
🧠 You Need to Know

📑 Watchdog Finds 332 Child Abuse Ads On Meta
→ The Tech Transparency Project reported that 332 ads containing AI-generated child sexual abuse material passed Meta's ad review and reached more than 29,000 accounts this year.

🧳 Anthropic Researcher Quits Over Superintelligence Race
→ Jacob Coxon, who spent three years on pretraining research at OpenAI and Anthropic, resigned Tuesday and said neither company is acting responsibly.

🚨 US Agencies Accuse Six Chinese Labs Of Distillation
→ A joint NSA, CISA and FBI advisory published Tuesday names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI as extracting billions of tokens from Claude, GPT, Gemini and Grok since late 2024.

🧠 Inception Ships Mercury 2.5 At 1,100 Tokens Per Second
→ Inception released the diffusion language model, claiming a 40% intelligence gain over Mercury 2 with a 260,000-token context window.

⚡ Google Puts $15 Billion Into Finnish AI Infrastructure
→ Alphabet said it will spend 13 billion euros across four Finnish cities in 2027 and 2028, including a 22-year Fortum nuclear agreement that is Google's first outside the United States.