📝🔍 Everything Claude Wrote for You This Month Has a Mark on It

Hey Explorers!

Anything Claude wrote for you this month carries a mark you can't see. McDonald's has 515 pages on one customer's order history. And an AI agent in Australia broke into a gym's booking system because someone asked it nicely. The record is being kept, whether you're the one keeping it or not….

Here's the signal that matters 👇🏼

📝 Your Claude Drafts Now Carry a Mark You Can't See
Anything Claude generated since August 2 has an invisible watermark baked into its word choices, and it survives copy and paste. It's global, with no opt-out, because the EU AI Act's Article 50 became enforceable that day. The wrinkle nobody planned for: a document you wrote yourself but had Claude heavily edit can still come back flagged, and the detection tool and dispute process don't exist yet. (🔗 Link)

🍟 McDonald's Has 515 Pages on One Guy's Order History
A Wired reporter requested his loyalty data and got back a file that went well past favorite menu items into predictions: how often he'd return, what he'd spend, and his odds of ever quitting, which the model put at zero. The program has 210 million active users and McDonald's CFO calls it the single most important digital metric in the company. Worth asking what your own loyalty data is predicting about your customers, and whether you'd want it read back to you. (🔗 Link)

🤖 Hand Off Work to a Bot the Way You'd Hand It to a Coworker
Grok Bot gives you a small team of agents with their own computer, signed into the tools you already use, working overnight and coming back only when something needs approval. A chief of staff bot sits on top with specialists underneath for inbox, expenses, recruiting, and ops, and they message each other in threads instead of routing everything through you. First product out of the SpaceX and Cursor merger, in beta on the upper tiers. (🔗 Link)

📰 Your Audience Data Is Worth More Than Your Traffic Now
USA Today's parent hired Palantir to build what its CEO called a "common intelligence layer" over reader behavior, where every visit and every session becomes a signal to monetize faster. Palantir has already run this play with Axel Springer and Thomson Reuters. If you're watching search traffic evaporate, this is the pivot the biggest newspaper chain in the country just made out loud. (🔗 Link) The Register

🔤 A Font That Feeds Scrapers Nonsense While You Read Normally
ShieldFont hides decoy words in your page's HTML, then uses OpenType substitution rules to swap the real ones back at render time. You read the sentence you wrote. A scraper pulling raw source gets about a quarter of the content words replaced with grammatically clean gibberish, though screen readers and copy-paste break too, and OCR defeats the whole thing. (🔗 Link)

🏋️ An AI Agent Hacked a Gym to Move Its Owner Up the Waitlist
An Australian developer asked his agent to bump him from spot four for a popular early morning class. It found an authorization hole in the booking software, cancelled the reservation of the person sitting in first place, and reported back like it had done him a favor. He was running Claude Opus 4.6, a model released in February, which is the part worth sitting with. (🔗 Link)

📬 Half Your Inbox Is Already Machines Talking to Machines
Gallup found more than half of US employees now use AI at work, and 51% of them use it for writing and editing, which means an agent-drafted email increasingly gets an agent-drafted reply. Banning it isn't the move. Name the three to five conversations whose value depends on a human actually showing up, declare those off limits, then give every agent a spending cap, limited credentials, and a named owner. (🔗 Link)

🎣 Your Agent Cheats Because You Rewarded It for Winning
When two OpenAI models broke out of their test sandbox and into Hugging Face in July, they weren't sabotaging anything. They were hunting for the answer key. Researchers call this reward hacking, and the hard part is that smarter models don't stop doing it. They just get better at hiding it. (🔗 Link)

The developer whose agent broke into the gym did the right thing afterward. He sent the company a responsible disclosure email laying out the vulnerability and suggesting fixes.

His agent wrote the email.

Somewhere in your business, a message is already going out that nobody actually wrote, and a file on a customer that nobody actually read. It’s really worth knowing which of those you'd put your name on.

These are the types of conversations we have with leadership teams. Where AI belongs in your operation, where it doesn't, and who owns the difference.

👉🏼 Work with Purple Horizons

Signal over noise. Always.

Ralph & Gianni

Researching accounts. Building lists. Writing sequences.

There's a better use of your team's time.

Apollo is the AI revenue engine that handles the busywork, so you can stay focused on selling.

Plus, everything you need is in one place:

Stop doing busywork and start building pipeline, faster.

With Apollo — the AI revenue engine powering 4M+ users.

Sign up for free

📍 9/1 South Miami Tech Tuesdays @ Bougies in South Miami

📍 9/8 Cafecito & Pastelitos @ Chug’s in Coconut Grove

Want to sponsor our newsletter or events? Drop us an email at [email protected]