How to stop a fake boss from moving your money, plus chatbot deal hunting, spotting fake stores, a Thanksgiving plan and a first AI errand.
New here? This is one weekly edition. If AI is new to you, start here first: one small task, ten minutes.
The call sounded like the boss. Prosecutors in New York say that was the point. Three men are charged with emailing accounting staff as their own chief executive, following up with calls that used AI to copy the CEO's voice and sometimes his face, then handing the employee to a fake lawyer at a big-name firm who arranged the wires. Prosecutors put the total at about $129 million, taken from hospital systems, a bank and a pet food company, among others. The men are charged, not convicted.
Not every target paid. According to the indictment, as reported by American Banker, a major New York investment bank's back office stopped a $2.85 million wire after an employee had already approved it. That's the lesson in one line: a second check that doesn't depend on who's on the phone. The same habit works on the family version of this scam, where the voice belongs to a loved one.
So, this weekend: set a call-back rule, let a chatbot hunt gift deals (then check every price), vet a store before you pay, get a Thanksgiving plan that knows how many ovens you have, and try an AI agent on one small errand.
The most useful things to try this weekend.
Federal prosecutors say three men used AI to mimic chief executives' voices, and sometimes their faces, on calls with accounting staff, then posed as outside lawyers to get about $129 million wired away. The defense is old-fashioned: hang up and call back on a number you already have.
In tests this fall, chatbots found real savings on holiday gifts and also quoted wrong prices, wrong models and fake 'deals'. Used as a scout, with every price checked, they can save you an afternoon.
One security firm linked about 119,000 web addresses to a single fake-shop operation that steals card details at checkout. Another found AI assistants recommending look-alike stores. A good-looking site proves nothing; a few minutes of checking makes you a much harder target.
Asked to plan Thanksgiving, AI is good at menus and timelines and shaky at actual shopping. In October, a grocery-industry editor got a 51-item, $202 list from one store's assistant and recipes that didn't match the list from another.
A Yahoo Finance reviewer let Meta's Muse agent get moving quotes, book dinner and buy a game. Two movers called within minutes. It also recommended a long-closed restaurant and admitted making up a phone number. Start with one small errand and keep the final click yours.
The shops are moving in. On November 10, Reuters reported that OpenAI now lets businesses claim profiles inside ChatGPT, add product feeds with prices, and run their own "business agents" that shoppers can chat with, starting with US online sellers. Businesses can also pay to promote those agents through ChatGPT's ads. OpenAI says sponsored placements are labelled and that it doesn't share your conversations with the businesses. In China, JD.com's shopping assistant drew nearly 80 million users during the Singles' Day festival, which ended November 11, and one analyst noted that some brands do better depending on how the platforms are tuned. Overall sales rose 4.8%, Reuters reported the next day, and nobody could say yet how much of that the chatbots actually sold.
In the tests so far, the bigger problem is plain accuracy, not salesmanship. In a study reported by 6abc Philadelphia in October, four popular chatbots answered 220 shopping questions and made costly mistakes, including a laptop price off by $300. A chatbot can be honest about its ads and still wrong about the price.
What to actually do: Use the chatbot to build the shortlist, not to choose. Ask it plainly which suggestions are sponsored, check the price on the store's own site, and remember that a store's own chat agent works for the store, however friendly it sounds.
Not every fake starts from nothing. Some start with a real, blurry picture that someone runs through an AI tool to "clean it up." After the November 1 jet crash in Palm Beach County (the one from last week's Slop Watch), a sharp still circulated on X asking whether the plane was missing a wing. Lead Stories found the image carried Google's invisible SynthID watermark, which flagged it as made or edited with AI. The real dashcam frames show both wings. The AI had invented the damage.
Old photos get the same treatment. In August, a colorized picture captioned as English schoolchildren saying grace before their morning milk in 1944 spread on X. The original is a real black-and-white photo in the Imperial War Museums' collection. Lead Stories compared the two and found the AI had redrawn the cups, sharpened a poster into letters that still don't read, and given the blurry children in the back row new faces. A real moment, with made-up details.
Copy this into ChatGPT, Claude or Gemini. Use roles ("owner," "my sister"), not names, and leave out account numbers and bank details. The AI doesn't need them.
A prompt is the instructions you give the AI. Anything in [square brackets] is a blank for you to fill in. First time? Start here
Help me write a short call-back rule for [my small business / my household]. The people who can send money are [roles, e.g. owner and bookkeeper / me and my spouse]. Include: 1) which payments or money requests always get confirmed by calling back on a number we already have, 2) the warning signs of a fake boss, lawyer, bank or family member (secrecy, urgency, new bank details, a new phone number), 3) a polite script for saying "I'll call you right back" to someone who sounds exactly like a person I trust, and 4) what to do in the first hour if money was sent by mistake. Keep it to one page in plain English.
Keep it private. Before you paste: leave out passwords, codes, and account, card, ID and policy numbers. Swap real names for "my mother" or "the company."
Rattled (7): “We are entering the age of recursive self-improvement (RSI) and no one knows what happens next.”
That's AI that helps build better AI, and plenty of researchers would agree with your second half. What we do know: the scams landing this week run on old tricks with new voices, and a call-back still beats them. We'll keep an eye on the big picture so your weekend doesn't have to.