
AI agents got cheaper and smarter this week. The fine print about who pays when they go wrong got a lot clearer too, and the answer is you. Google shipped a faster, lower-priced model built for exactly this kind of agent work, while some of the biggest banks in the country rewrote their terms so that an AI agent's mistakes land on the customer. Let's get into it.
Stop feeding the feed.
You’ve put in the work to build an audience. But when the only way to reach your fans is through a social feed determined by algorithms, the platform decides who sees what you make next.
beehiiv gives you the tools to reach your audience directly. On their all-in-one platform, you can turn followers into subscribers, grow your newsletter, and connect it all to your community or podcast. Stop chasing virality on platforms you can’t control. Build a stronger, more sustainable business on beehiiv.
Google's New Mid-Tier AI Is Nipping at the Leaders
Google released Gemini 4 Neon, a faster model aimed at coding, office work, and other multi-step "agent" tasks. According to Google's announcement, it scored 61 on the Artificial Analysis Intelligence Index, good for third place overall, and ranked first on that group's coding agent index. Business pricing through the API is $3 per million input tokens and $15 per million output tokens. It's live now for Google AI Pro and Ultra subscribers in the Gemini app, and for businesses through the Gemini API.
Why it matters to you: Skip the benchmark bragging, the part that matters is price for performance. Near-top-tier AI keeps getting cheaper, which means the tools built on top of it (your CRM's assistant, your website chatbot, your writing tool) should keep getting better without costing you more. Keep in mind these are Google's numbers, and the only test that counts is how it handles your actual work.
What to do about it: If you already pay for Google AI Pro, run one real task you'd normally give ChatGPT or Claude through Neon and compare the results side by side.
Your Bank Says AI Agent Losses Are on You
Citi updated its online user agreement this week to say customers who give third-party AI agents access to their accounts are "responsible for activities conducted through those agents," TechCrunch reports. Per the report, that includes what a hacker does with your account if the AI tool leaks your login or gets breached. PNC changed its terms the same day, Citizens emailed customers about it, and JPMorgan and Bank of America announced similar policies last month. Citi's own in-house AI agents don't come with that extra liability.
Why it matters to you: Every AI vendor is pitching agents that "handle your finances" or "reconcile your books." The banks just told you, in writing, who's accountable when one of those goes sideways. On a business account, that's real money. An agent that can only read your transactions is a very different risk from one that can move money.
What to do about it: Before connecting any AI tool to a business bank account, ask whether it needs to move money or only read data, and choose read-only whenever you can.
Most AI events send you home with a notebook full of ideas and no time to use them. This one sends you home with the thing already built.
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 don't leave with a framework you still have to go implement. You leave with a finished asset that's ready to work for you. It's built for business owners who are tired of running their sales process on their own time and energy and want it running as a system instead. Last year's event sold out, and it's coming back with a bigger AI focus. I wasn't there last year and I had major FOMO about it, because I know a lot of the trainers, and they weren't talking theory, they'd done the work. I'm going this 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
A Federal Watchdog Got 4 Big Chatbots to Write Attack Plans
The Government Accountability Office publicly released a report Wednesday saying it got ChatGPT, Claude, Gemini, and Grok to produce attack plans its experts judged workable, according to GAO. The tests ran in spring 2026, using two scenarios: a mass-casualty explosive device and an attack on the power grid. Every chatbot's safety guardrails kicked in at first, but GAO got past them using "common safeguard-bypassing techniques," and each produced a plan in under 12 hours. The report labels the chatbots A through D and doesn't say which one produced which plan.
Why it matters to you: This won't change how you use ChatGPT to write an email. But it's a clear reminder that AI guardrails are a speed bump, not a wall. The same goes for any chatbot you put in front of customers: if it can be talked into saying something it shouldn't, eventually someone will try.
What to do about it: If you run an AI chatbot on your website, spend 15 minutes trying to get it to misbehave (promise a discount, trash a competitor, make up a policy) before a customer does it for you.
Microsoft's AI Agent Toolkit Was Hijacked to Steal Passwords
On November 17, two poisoned versions of agent-framework, Microsoft's official toolkit for building AI agents, went out through Microsoft's own publishing channel, according to security firm Wiz. Versions 1.26.1 and 1.26.2 quietly stole cloud keys, passwords, and other credentials from the computers that installed them. Both were pulled within about an hour, but Wiz found three confirmed cases where victim data was stolen. Wiz says the toolkit gets roughly 41,000 downloads a week.
Why it matters to you: You probably don't install developer toolkits yourself. But if a developer, agency, or freelancer builds AI tools for your business, they very well might. Software from the biggest names in tech can be compromised right at the source, and a stolen cloud key can turn into stolen customer data.
What to do about it: If anyone builds AI tools for you, send them one line: "Did we install agent-framework 1.26.1 or 1.26.2? If so, please rotate every key."
Washington May Make Businesses Admit They Use Chinese AI
The Trump administration is weighing a rule that would force U.S. companies to publicly disclose on their websites when they use Chinese AI models, Reuters reported. Treasury and the White House science office have discussed the idea, though Reuters says it's unclear whether it's gaining traction or would become a formal rule. Airbnb, which uses Alibaba's Qwen model for its customer service agent, is cited as an example, and two House Republicans have sent letters to Airbnb, DoorDash, and Anysphere, the company behind the coding tool Cursor. Chinese models were the three most-used models on the developer marketplace OpenRouter in October.
Why it matters to you: Plenty of low-cost AI tools quietly run on Chinese models under the hood, because they're cheap and good. If this becomes a rule, businesses could be expected to know what's powering the software they use. Nothing is required yet, so there's no reason to panic, but it's worth knowing what you're running.
What to do about it: Next time you sign up for an AI tool, ask the vendor which model it runs on. A trustworthy vendor will just tell you.
The best business decision I've made in the last 3 years wasn't a tool or a hire. It was a room.
Ignite Mastermind is an annual membership built for entrepreneurs still under $1 million a year who want structure, tools, and a group of peers to get past that milestone faster. 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. Here's what comes with it: 6 to 8 live 3-day virtual workshops a year, tickets to 2 in-person multi-day events, weekly mentorship calls with Perry Belcher and his team, and an all-in-one marketing software suite that replaces over $5,000 a month in separate tools. Each of those workshops and events normally runs around $97 to $997 on its own. And as my personal bonus, you get monthly accountability and business development calls with me. If you're looking for a mastermind to move you to your next milestone, this is where I'd point you. See if Ignite is right for you
ChatGPT Now Wants to Be Your Sales Assistant
OpenAI launched ChatGPT for Sales, a plugin with built-in workflows for prioritizing accounts, prepping for calls, drafting follow-ups, and flagging deals at risk. It comes with more than 30 new and upgraded connections, including Salesforce, HubSpot, Close, monday CRM, Gong, Granola, Docusign, and PandaDoc. It's available to ChatGPT Business and Enterprise workspaces, where an admin switches it on. OpenAI didn't list a separate price and says availability varies by plan and region.
Why it matters to you: If your pipeline already lives in HubSpot or Salesforce and you pay for ChatGPT Business, this could replace some of the "AI sales assistant" add-ons you're getting pitched every week. The catch: it's only as good as your CRM. AI can't prep you for a call using notes nobody ever wrote down.
What to do about it: If you're on ChatGPT Business, have your admin enable it and try it on one call prep before you pay for a separate sales AI tool.
An AI Detector Called a Prize-Winning Novel 99% Fake
The AI detector Pangram flagged 99% of "Donde el agua trabaja," an award-winning novel by Peruvian author David Silva Rojas, as machine-written, The Decoder reports. Per El País, it also flagged more than 80% of his interview answers, an essay, and even the Instagram post where he denied using AI. His Mexican publisher, Almadía, reviewed documents from the author and is standing by him. Meanwhile, French publisher Gallimard reportedly plans to start screening manuscripts with Pangram next year.
Why it matters to you: AI detectors produce estimates, not verdicts, and the Authors Guild warns against using one as the only basis for a decision. If you're tempted to run a job applicant's writing sample or a freelancer's work through a detector and act on the score alone, this is your reason not to.
What to do about it: Treat a detector score as a reason to ask questions, never as proof.
Google's AI Search Now Has an Incognito Mode
Google added "temporary threads" to AI Mode in Search, TechCrunch reports. Conversations in a temporary thread aren't saved to your search history, aren't used for recommendations, and don't shape your future personalization. You start one by tapping the temporary threads icon in the upper right corner of AI Mode, no settings change needed. Google says it's rolling out globally on desktop and mobile.
Why it matters to you: Business owners ask AI search about sensitive things all the time: a competitor, an employee situation, a legal question. Now those one-off questions won't follow you around in your recommendations. Just know that keeping something out of your profile isn't the same as keeping it from Google entirely.
The Bottom Line
Here's today's pattern. The tools keep getting cheaper and smarter, and Google proved it again. But the fine print is moving just as fast, and it's moving in one direction: toward you. Your bank won't cover an agent's mistakes. A Microsoft toolkit got poisoned at the source. A federal watchdog talked 4 chatbots past their own guardrails. None of that means stay away from AI. It means stop treating an "AI agent" like a magic employee and start treating it like a new hire with access to your accounts: limited permissions, someone checking the work, and nothing it can do that you can't undo. That's how you get the upside without getting thrown off the ride.
Enjoying the Ride?
If this issue helped you make sense of the week, forward it to one business owner who's trying to keep up with AI without losing their mind. And if someone sent this to you, subscribe so you get the next one straight to 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




