Anthropic Says AI May Soon Copy Itself Onto Cheap Gadgets. The Catch: Its Test Made Hacking Easy.

Today's stories share one thing: the gap between the headline and the fine print. Anthropic warns that AI could copy itself onto cheap gadgets, using a test it made easier on purpose. Epic Games told a team to stop writing code by hand. Publishers that police authors for AI are reportedly using it themselves.

The best person for your next role might not live near your office — or even in the same country.

And more companies are realizing they don't need to open entities everywhere just to hire great talent globally.

Instead, teams are using an employer of record (EOR) to hire internationally faster, stay compliant, and avoid the operational headache of setting up local infrastructure before they're ready.

That shift is changing how companies think about growth, hiring, and expansion altogether.

Oyster's EOR helps companies hire, pay, and support employees in 180+ countries while Oyster handles payroll, compliance, taxes, and local employment requirements.

Talk to experts today

Anthropic published research on whether small AI models could copy themselves onto cheap devices. Models of 30 billion parameters or fewer could complete many individual steps, like setting up software on unfamiliar hardware, but often failed when they had to chain the steps together. One open model, GLM-5.2, copied itself across a simulated 8-device network in 4 of 5 trials, though the researchers deliberately made the hacking step trivial by using an unprotected port. Anthropic's headline finding is that small models may be only months away from being capable enough to do this, but the authors call their results an upper bound, and even the working setups took roughly 28 to 150 hours.

Why it matters to you: This is a warning aimed at security researchers and internet providers, not a reason to unplug anything. The weak spot it describes is cheap connected gear with default passwords or open ports, like off-brand TV boxes, cameras, and routers. The forecast is a trend line, not a finding, so treat it as a reason to tidy up basics rather than a reason to worry.

What to do about it: Change the default password and update the firmware on every camera, router, and smart device at your business.

Bloomberg reported, and Engadget summarized, that a director told the team behind Epic's Disney-related game to go "AI first." The internal message reportedly said "No more hand-written code" and ordered that all art be made with full AI workflows. The game, rumored to be an extraction shooter, was reportedly pushed to March before that order came down. Epic says it is pursuing "AI-accelerated development" led by its own developers, while Engadget notes a GDC survey in which 52% of developers say generative AI is bad for the industry, up from 30% in 2025.

Why it matters to you: Epic's delay reportedly came before the AI order, so the robots can't take the blame, and a mandate from the top doesn't make work faster by itself. Some staff reportedly hoped the order wasn't serious, which is what happens when a rollout is announced instead of tested. The same trap waits for any small team told to "use AI for everything."

What to do about it: Pick one repeating task, run it both ways for two weeks, and compare time and errors before you mandate anything.

Most business owners I know are great at what they sell and worn out by how they sell it. Every lead, follow-up, and proposal runs on their own time and energy. That is the problem the Epic Marketing Summit goes after, and it does it by building instead of lecturing.

It is a 2-day live event in Miami where you build a working AI-powered sales funnel and vibe code your own AI app on-site. You walk out with a finished asset, not a binder of frameworks you still have to go implement. Last year's event sold out, and it is 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. Grab your spot here

Character.AI started a voluntary Chapter 11 process to run a court-supervised sale of the company, citing legal, regulatory, and operational headwinds. According to PPC Land's breakdown of the filing, Disney is the opening bidder at $20 million for a non-exclusive, perpetual license to certain company IP, and that offer does not include the app itself. Rival bids can top it. The company says that if no buyer agrees to keep the platform running, it will be wound down, with the sale expected to close around late November or early December.

Why it matters to you: Character.AI's service and paid subscriptions are still running for now, but this is a reminder that AI vendors can vanish fast, even well-known ones. If a chatbot or AI tool sits in front of your customers, the company behind it is part of your risk.

What to do about it: For any AI tool you depend on, confirm you can export your data and name a backup option.

Anthropic said new defenses have cut the data harvested by "distillers," meaning companies that use Claude's answers to train copycat models, by around 90 to 99%. It calls that early evidence and admits it doesn't know how long the gains will last. The defenses include summarizing Claude's reasoning and a classifier that spots extraction attempts. Since September 24, Anthropic also charges for requests its safeguards block before Claude responds, and that includes the rare innocent request caught by mistake, which it puts below 0.1%.

Why it matters to you: This only touches you if you or a vendor build on Claude's developer connection, not if you just use the chat app. But if someone built an automation for your business, odd charges for requests that never got an answer could show up on a bill.

What to do about it: Ask whoever built your AI automations to check the billing logs for charges on blocked requests.

X's pricing page lists a $30 a month Premium tier with about 4 times the usage of its $8 Plus tier, plus Grok Bot, code hosting, and cloud agents. Usage comes from one shared pool across Grok products that resets weekly, and you can buy top-ups if you run out. Other companies' models used inside Cursor are not included. They are billed separately at API list prices, though you can set spending caps. Prices exclude taxes.

Why it matters to you: A bundle price only helps if the bundle covers what you actually use. If most of your work runs on Claude or ChatGPT, the $30 sticker leaves out your biggest line item.

What to do about it: Before switching any AI plan, write down which model you use most and price that one first.

The fastest way I know to stop growing alone is to sit in the right room.

Ignite Mastermind is an annual membership for entrepreneurs still under $1 million a year in revenue who want to get past it faster. You get 6 to 8 live virtual workshops a year, tickets to 2 in-person multi-day live events, and weekly mentorship calls with Perry Belcher and his team. Those workshops sell on their own for around $97 to $997 each. It also includes an all-in-one marketing software suite that replaces over $5,000 a month in separate tools, all connected, plus a peer community built for real partnerships.

I've been a member for over 3 years, and it's been a vital part of how I've grown my business. Some of my closest friends and a few real business partners came directly out of being in that room. I also add a personal bonus: monthly accountability and business development calls with me. If you want a mastermind to carry you to your next milestone, this is where I'd point you. See if Ignite is your room

WIRED spoke with more than two dozen anonymous publishing employees, and Gizmodo's summary says three of the "big five" publishers are using AI without telling authors or the public. The reported uses include ChatGPT and Claude for publicity and cover copy, image generators for cover art, expanded translation work, and monitoring employee productivity. One agent says publishers now agree not to run unpublished manuscripts through AI tools for promotional text, a promise meant to protect authors from tech companies that also binds the publisher itself. Gizmodo doesn't name the publishers, and the sourcing is anonymous, so treat the specifics as unconfirmed.

Why it matters to you: Publishers sell human writing, so quietly using AI on the copy that sells it is a trust problem in waiting. Your customers may feel the same way if they learn AI wrote your emails or ads and you never said so.

What to do about it: Decide your AI disclosure rule now, even if it is one sentence on your website.

Every big claim today came with a footnote. Anthropic's AI worm warning rests on a test it made easy. Epic's AI order landed after a delay, so the AI gets neither the blame nor the credit. X's $30 bundle leaves out the models you may actually use, and Character.AI shows that even a famous AI product can end up in a courtroom sale. Nothing here needs a panic or a purchase. It needs you to read the fine print before you act. Stop the rollercoaster, check the footnote, then decide.

Forward this issue to one person who is tired of AI hype. If someone forwarded it to you, you can subscribe free at stoptheairollercoaster.com.

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