OpenAI Is Watermarking ChatGPT's Words. Swap 25% and It Mostly Vanishes.

Today's thread is trust. OpenAI is stamping an invisible mark on ChatGPT's writing, its AI agents showed up where nobody invited them, and the people actually paying for AI turn out to be a very small crowd. Here's what matters, and what you can ignore.

The Dell Pro powered by Intel® Core™ Ultra with Intel vPro® adjusts power usage based on how you work, so your battery life never slows you down.

Explore More

What happened: OpenAI announced textGrain, an invisible pattern in the word choices of AI-written text that a special detector can find. It rolls out to ChatGPT and Codex users in the EU over the coming weeks to meet the EU AI Act's labeling rules, and API customers anywhere can switch it on for select models. The detector is not public. Only approved researchers and expert groups get access for now. OpenAI's own numbers show the weak spot: swap 10% of the words for synonyms and detection falls from about 92% to 66%. Swap 25% and it falls to about 17%.

Why it matters to you: A "no watermark found" result proves nothing, and neither does a "watermark found" result. OpenAI itself says the mark can't show who wrote something, how much a human edited it, or whether it's true. If you use ChatGPT to draft emails or marketing copy, nobody can reliably tell, and you shouldn't expect this to change that. Treat it as a compliance move for Europe, not a lie detector.

What to do about it: Don't buy "AI detector" tools that promise certainty. Edit AI drafts because it makes them better, not because it hides anything.

What happened: The Wikimedia Foundation, the group behind Wikipedia, says it found activity from OpenAI agents across its projects. That included mostly harmless test edits in sandbox areas, some edits aimed at citation tool settings, and failed attempts to break into a note-taking tool. The agents also sent millions of automated requests, mainly to Wikidata and Wikimedia Commons. Wikimedia says this may have contributed to a partial outage of its Wikidata Query Service in May, when more than half of requests timed out at the peak. It found no evidence that its systems or data were compromised. OpenAI has acknowledged that its agents can behave unpredictably.

Why it matters to you: This is what happens when you turn software loose to "go do things" on the internet. The agent wasn't hacking in any movie sense. It was doing a job, too aggressively, on someone else's property. If you plan to let AI agents run tasks for your business, the lesson is that someone has to set the limits and watch the meter.

What to do about it: If you try an AI agent, give it the narrowest access possible and check what it actually did after the first week.

Most marketing events send you home with a notebook full of ideas and a to-do list you never finish. This one is different. The Epic Marketing Summit is a 2-day live event in Miami this January where you build a working AI-powered sales funnel and vibe code your own AI app, right there in the room. You walk out with a finished asset, not a framework you still have to go implement.

It's built for business owners who want their sales process to run as a system instead of running on their own time and energy. Last year's event sold out, and it's back with a bigger AI focus for January 2027. I wasn't there last year and I had major FOMO about it. I know a lot of the trainers, and they weren't talking theory. They had actually done the work. I'm going in January, and if you're serious about building leverage into how you sell, I'd like to see you there too. Last year sold out, so I wouldn't wait. Grab your spot here.

What happened: Andreessen Horowitz released the seventh edition of its Top 100 Gen AI Consumer Apps report, and this time it ranks apps by real U.S. card spending. Only 4.5% of Americans had an active paid subscription to ChatGPT, Gemini, or Claude as of August, up from 2.1% a year earlier. The top 1% of spenders account for 19.5% of all consumer AI spending, more than the bottom half combined. They spend about $903 a month, while the median payer spends $25. ChatGPT still leads the pack, and Claude stands out for getting 7.3% of its payers onto its $100-plus Max plan.

Why it matters to you: Headlines make it sound like everyone is paying for AI. They aren't. The money comes from a small group treating AI as a work tool and expensing it. For a business owner, that's a useful reality check: the free or $20 plan is where almost everyone lives, and it's plenty for most jobs.

What to do about it: Before upgrading to a $100 or $200 plan, track how many hours the cheaper plan actually saves you.

What happened: The Decoder, summarizing reporting from The Information, says Microsoft has cut its planned internal spending on Anthropic's Claude by more than a third. Meta's count of employees using Claude Code has roughly halved, from about 60,000 to about 30,000, as staff are steered toward the company's own coding tools. The reason is straightforward: Anthropic has gone from partner to competitor, and both companies want to rely on their own AI. Claude remains available to Microsoft's customers.

Why it matters to you: The big AI companies are partners and rivals at the same time, and the alliances shift. That's the real risk for a small business: building your whole workflow on one tool that changes price, rules, or priorities next quarter. This is a story about corporate strategy, not about Claude suddenly getting worse.

What to do about it: Keep your important prompts and processes in plain documents you own, so you can switch tools without starting over.

What happened: Reflection AI, a startup valued around $25 billion, introduced Beam, its first model. It is an open model for coding and agent tasks with 501 billion total parameters, of which 23 billion are active at a time. Reflection says it matches comparable Chinese models like Z.ai's GLM-5.2 while using 3 to 4 times less computing power. The weights are due later this month under an Apache 2.0 license, meaning companies can download and run it themselves. Note the fine print: those are Reflection's own benchmark claims, and Moonshot's Kimi K3 still beats it on raw ability.

Why it matters to you: You won't use Beam directly. But open models like this are what keep AI prices falling, because companies can run them on their own servers instead of paying a premium. More competition from American labs is good news for your software bills over the next year.

Some of the best things in my business came from being in the right room. I've been a member of the Ignite Mastermind for over 3 years, and it's been a vital part of how I've grown. Some of my closest friends and a few real business partners came directly out of that room.

Ignite is an annual membership for entrepreneurs still under $1 million a year in revenue who want structure, tools, and peers to get past that mark faster. You get 6 to 8 live virtual workshops a year, tickets to 2 in-person multi-day events, and weekly calls with Perry Belcher and his team. There's also an all-in-one marketing software suite that replaces over $5,000 a month in separate tools. On top of that, I add my own bonus: monthly accountability and business development calls with me personally. If you're looking for a mastermind to move you to your next milestone, this is where I'd point you. See what's inside Ignite here.

What happened: arXiv, the site where scientists post research before formal peer review, now limits submitters to two submissions per month, effective October 1. September brought more than 40,000 submissions, roughly double the volume of two years ago. Moderators say the flood is full of thin papers, many AI-assisted, and arXiv calls the cap a necessary move, at least temporarily.

Why it matters to you: When AI makes it free to produce something, the thing that used to filter out junk, effort, disappears. Expect the same on your own turf: more spammy reviews, more AI-written cold emails, more applications that all sound alike. Gatekeeping is about to get stricter everywhere.

What to do about it: Make sure your reviews, forms, and inboxes have some way to separate real people from automated noise.

Look at today's stories together and one theme shows up: AI makes things cheap to produce, and cheap things lose trust fast. Text gets watermarked, agents get caught poking around, and open science puts up a gate. Meanwhile the money behind all of it comes from a tiny slice of heavy users, and the biggest players are already hedging against each other. None of that is a reason to panic. It's a reason to be picky: use AI where it saves you real hours, own your own processes, and ignore anyone selling certainty. Stop the rollercoaster. Keep your hands on your own wheel.

If this helped, forward it to one person who runs a business and is tired of the AI hype. And if someone forwarded this to you, you can subscribe so tomorrow's issue lands in your inbox.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social