🍭 AI Is Watermarking Molecules 🧬

Good morning.

- “Sir, this protein has no watermark."
- "It's an egg."

Let’s dive in 👇

🍭 What’s Cookin’:

• Google started watermarking proteins. Yes, the molecules.

• OpenAI got sued over the agents that hacked Hugging Face

• The FTC is now looking at OpenAI and Anthropic too

Google DeepMind
🧬 Put A Watermark On Life Itself

The Bite:

Google DeepMind unveiled SynthID Bio on September 30.

The system hides a detectable signature inside AI-designed proteins.

The signature sits in the protein's amino acid sequence and in the atomic coordinates of its predicted 3D structure.

DeepMind says the signature can be detected on the physical protein after it is made.

The tool is aimed at DNA synthesis providers, that check DNA orders for potentially dangerous sequences before making them.

DeepMind is publishing the method in Nature, open-sourcing the code and releasing the model weights to researchers.

Snacks:

• Wet-lab tests covered three targets, including the spike protein of the COVID-19 virus.

• Watermarked designs matched unwatermarked ones on hit rate, binding strength and sequence diversity.

• DeepMind fine-tuned part of AlphaFold 3, so every structure it predicts carries the signature.

• Resistance to deliberate tampering is still listed as an unsolved challenge.

• With Stanford's Hie lab and Arc Institute, DeepMind watermarked the genome of an AI-designed virus that infects bacteria.

• Early lab tests show the watermarked viruses still work.

Why it Bites:

Let's take a moment here: Google is watermarking molecules now.

Not essays. Not deepfakes. Proteins, the microscopic machinery every living thing runs on. An AI designs one, a lab builds it, and somewhere in the amino acids sits a tiny "made by Google."

Genuinely clever science. And the wildest part?

Google just built the compliance tool for a rule that doesn't exist yet.
Which means the rule can exist now.

Until this week, "label every AI-designed molecule" sounded like sci-fi. Now it's open-source code with a Nature paper behind it, and any regulator who wants to mandate it has a working example to point at.

Now regulators just have to move faster than the labs.

Ideally before someone founds Umbrella.

Steal This Prompt
🏙️ Your Design → Giant City Billboard

Take your boring flat design and drop it into the real world, looking suspiciously expensive. Upload your plain design, pick a country, and let AI slap it onto a massive modern-city billboard while keeping the original design intact.

Use it to:

• Turn brand designs into realistic advertising mockups

• Show clients what a campaign could look like IRL

• Create portfolio visuals that say “yeah, we have a billboard” without actually buying one

Workflow:

1. Hit this link: Your Product On a Billboard

2. Upload your plain design alongside the prompt

3. Replace the #’s with your country + extra scene details

4. “wait… did you actually rent that billboard?”

   Try on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

✈️ Cura

Plans the trip, picks flights and stays with its reasoning shown, handles check-ins and changes, and hands you to a human advisor when things get messy.

🔧 Yedric.ai

Drop-in agent that lets your SaaS users just type what they want done, wired to actions your product already performs, live in under 30 minutes.

🤖 Dots by OpenAI

Always-on agents that each get their own cloud computer and browser, plug into 4,000+ apps, and take calls on Slack, Teams, or your phone.

💻 DSH Desktop - By Docker

DeepSeek's open-source agent harness as a Mac and Windows app, running your coding, everyday work, and scheduled tasks locally.

🔍 Omnia Agent

Finds the AI search prompts where competitors show up instead of you, then goes after them, with every move waiting for your approval.

_{Everything Else}
🧠 You Need to Know

📜 Newsom Signs Laws Shielding Workers From AI
→ The California laws bar employers from relying on AI to fire workers or to read their emotions from biometric data, and require written notice when AI drives mass layoffs.

⚖️ OpenAI Sued Over Its Agents' Hugging Face Hack
→ Nonprofit Legal Advocates for Safe Science and Technology is seeking a court order barring OpenAI's agents from accessing computer systems without authorization.

💸 ElevenLabs Doubles Its Valuation To $22 Billion
→ A $300 million employee tender offer closed Wednesday values the voice AI company at twice the $11 billion it reached in its February funding round.

🔍 FTC Opens A Probe Into OpenAI And Anthropic
→ The Federal Trade Commission confirmed it is investigating OpenAI, Anthropic and other AI companies over the risks their technology may pose to consumers.

🧬 Google DeepMind Launches A Watermark For AI-Designed Proteins
→ Unveiled Wednesday, SynthID Bio embeds a detectable signature in AI-designed proteins, and in wet-lab tests the watermarked proteins worked as well as unmarked ones.