This AI Fooled 48% of People on Video. The Fix Isn't Ready.

Today's thread is trust. An AI convinced nearly half the people who met it on video that it was human. OpenAI parted ways with three safety researchers and has said very little about why. And a Meta shopping agent handed a stranger a home address because its owner clicked "Allow Always." Here's what happened and what to do about it.

What does it look like when a consumer brand starts to break out?

Popsmith generated $5.8M in revenue in 2025 and is projecting $9M in 2026, with more than 100,000 customers and $20M+ in lifetime sales.

The brand has already earned national recognition through Oprah’s Favorite Things and partnerships with Costco, Sur La Table, and Williams Sonoma, where Popsmith is rolling out nationwide this holiday season.

Now Popsmith is raising capital to accelerate its next stage of growth, and individual investors have the opportunity to become shareholders.

Explore the Popsmith investment opportunity.

What happened: Tavus, an AI video startup, previewed a model called Griffin that watches and listens during a live video call while it talks, so it can nod mid-sentence, interrupt, and react to what it sees. In a company-run study, 26 of 54 people, or 48%, believed they had met a real person after a one-minute call. The company's previous model scored 2.4%. Griffin-Lite is limited to trusted testers for now, and Tavus says it is still building disclosure features, with no timeline given. Keep in mind the study was small and run by Tavus itself.

Why it matters to you: A video call has been the last easy way to confirm someone is real. If this tech spreads, "I saw them on camera" stops being proof, which is a gift to scammers faking a vendor, a job candidate, or a boss. Nothing is available to the public yet, so treat this as a heads-up, not an emergency.

What to do about it: Set a code word or a call-back number for anyone who asks you to move money or share account access over video.

What happened: OpenAI confirmed it let go of three people in its safety organization after an internal investigation found they mishandled sensitive company information outside established procedures, TechCrunch reports, following a Wall Street Journal story that the information went to an outside AI safety group. OpenAI has not named the group, the information, or the people. The news lands after reports of AI agents escaping containment and OpenAI scrapping its planned GPT-6.1 Astra launch over safety concerns.

Why it matters to you: The missing details are the story. From the outside, nobody can tell whether this was a real leak or a company squeezing out people who raised alarms, and OpenAI isn't saying. You don't need to panic. You do need to avoid building anything critical on a single vendor.

What to do about it: Keep one backup AI tool you could switch to in an afternoon.

Most business owners I know don't have a sales problem. They have a sales process that lives in their head and runs on their energy. That's why I'm pointing you at the Epic Marketing Summit in Miami this January. It's a 2-day live event where you don't sit and take notes. You build. Over two days you put together a working AI-powered sales funnel and vibe code your own AI app on-site, so you walk out with a finished asset instead of a binder of ideas you still have to go implement. It's made for owners who want their sales process to run like a system, not a second job. Last year's event sold out, and it's back with a bigger AI focus. I wasn't there last year and I had major FOMO. I know a lot of the trainers, and they weren't teaching theory. They had actually 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

What happened: Meta's Muse agent negotiated the sale of a keyboard on Facebook Marketplace for tech reviewer Matt Robb, accepted a lower offer, shared his pickup address with the buyer, and then messaged "Yup, I'm here!" when the buyer showed up at his building around 9:15 p.m., TechRepublic reports. Robb had chosen "Allow Always," which he believed would still require approval for consequential actions. Meta says there was no breach of privacy controls and that it plans to make permission prompts clearer.

Why it matters to you: An AI that acts for you will read "always" more broadly than you expect. The settings screen and the AI's idea of what you agreed to can be two different things.

What to do about it: Choose "ask me each time" for anything involving payments, addresses, or customer data.

What happened: Governor Newsom signed a batch of AI bills on September 30, according to the governor's office. SB 947 calls for real-person review before an employer disciplines or fires someone based on an automated system, so AI can't be the only decision-maker. SB 951 requires notice when mass layoffs, relocations, or terminations involve AI systems. Two more bills restrict workplace surveillance tools, including in bathrooms. The announcement did not list effective dates.

Why it matters to you: California tends to set patterns other states copy. If you use software to screen applicants, schedule shifts, or flag low performers, a human should make the final call, and you should be able to say who did. That's smart practice in any state.

What to do about it: Check whether any HR or hiring tool you use makes pass or fail decisions on its own.

What happened: Anthropic's IPO filing shows Broadcom agreed to lend up to $42 billion in convertible notes, roughly a third of Anthropic's $125.2 billion, five-year commitment for TPU chip capacity, Startup Fortune reports. Broadcom co-designs those chips with Google, so the supplier is financing the customer's purchase of the supplier's product. The filing also warns that a payment default could speed up lease obligations while limiting access to the loan. It lists about $518 billion in total infrastructure obligations against $4.6 billion in revenue and more than $8 billion in losses.

Why it matters to you: AI companies are funding their growth with enormous debt, often owed to the companies selling them the equipment. That doesn't threaten your subscription tomorrow. It does mean prices, limits, and availability depend on lenders staying friendly.

What to do about it: Think twice before prepaying a long contract for any single AI tool.

The best move I ever made for my business wasn't a tool or a tactic. It was getting into a room with people a few steps ahead of me who would tell me the truth. 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. Here's what's inside: an annual membership with 6 to 8 live 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. It's built for entrepreneurs still under $1 million a year in revenue who want structure, tools, and peers to get past that milestone faster. My personal bonus is monthly accountability and business development calls with me. No pressure from me. If you're looking for a mastermind to move you to your next milestone, this is where I'd point you. See how Ignite works here

What happened: Reddit says new public API requests end October 31, RSS feeds stop working November 13, and API access closes fully by March 2027, as part of an effort to make scraping harder, Decrypt reports. Third-party apps, moderation tools, and reminder bots take the hit. Reddit is offering a $1 million migration program to move developers onto its own Devvit platform.

Why it matters to you: If you or a vendor use Reddit feeds to watch for mentions of your brand or your competitors, those alerts will quietly go silent on November 13. It's also one more sign that open web data is getting locked down as companies fight AI scrapers.

What to do about it: Ask any monitoring tool that pulls from Reddit what its plan is before the cutoff.

What happened: Google announced Gemini 4 Argon, which it says can write up to 1 million tokens in a single run and leads several benchmarks, including 77.9% on DeepSWE v1.1. Access is limited to vetted cyber defenders for now, with paid API customers and Google AI Ultra subscribers next, at an intro price of $2 per million input tokens and $10 per million output, rising later to $4 and $20. Bloomberg, as relayed by Android Headlines, reports some Google employees find it strong on benchmarks but weaker on real-world coding and web design. Google DeepMind leadership disputes any gap.

Why it matters to you: Benchmarks are the maker's own test scores. Wait for people doing work like yours to report back before you switch tools. You can't get in yet anyway.

Look at the pattern. An AI that passes as human. An agent that shared an address because someone clicked "always." A lab that parted ways with three safety researchers and told us almost nothing. None of it calls for panic. All of it says the same thing: the technology is moving faster than the guardrails, and in your business, you are the guardrail. Verify the person, limit the permission, keep a backup. Stop the rollercoaster, get off at the stops that matter, and keep one hand on the bar.

If today saved you from one headline you would have missed, forward this to one person who runs a business. And if someone forwarded this to you, you can subscribe here 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