Sam Altman said the quiet part on the record this weekend. Asked where he disagrees with Anthropic, the OpenAI chief told POLITICO that "the world should accept some bad things happening" in exchange for what AI can do. He said he would not take a deal that promised zero scams if the price was locking the technology away.
You can agree with him or not. I'm not here to referee a fight between two labs. I'm here because the word he used was scams, and scams are where online sellers live and bleed. So today's digest is about one theme: the people building and governing AI have told you, fairly plainly, how much protection to expect. Plan for that number, not a bigger one.
Altman won't promise you zero scams
In the interview with POLITICO's Decoded, Altman drew a line between OpenAI and Anthropic. His version: one camp thinks AI will get so dangerous that a single lab should hold it and hand out the benefits carefully; his camp thinks people should have broad access, and that society should accept some harm along the way. He called the zero-scams trade-off "completely unacceptable". He did say he would not accept catastrophic risk, such as a serious loss of control to AI.
Worth knowing: the gap between the two companies is narrower than the headline. OpenAI has recently backed some tougher state safety laws and a bipartisan proposal for independent model testing. But the message to everyday users is clear enough. The tools will stay open, and some people will use them to cheat other people.
Why it matters for sellers: you're the "some people" on the receiving end. Fake bank receipts that look perfect. Voice notes that sound like your supplier. Screenshots of a "transfer" that never happened. Two days ago we covered how a cloned voice and a fake WhatsApp account drained €95 million from a bank; a small shop has far fewer people checking than a bank does. So build your own floor:
- Never ship on a screenshot. Check the money in your actual account.
- Agree a code word with suppliers for any change in bank details.
- Treat "urgent, send now" from any voice or number as a red flag, not a reason to hurry.
Washington builds a "Super Intelligence Force", and the FTC is in the room
This weekend, President Trump named the leaders of a new federal AI task force he's calling the Super Intelligence Force. It's chaired by Director of National Intelligence Jay Clayton, with FTC Chair Andrew Ferguson, Pentagon research chief Emil Michael and personnel-office director Scott Kupor. It reports to the president and his chief of staff, and has 120 days to deliver a report on AI's risks and benefits and on what the government's role should be.
This comes days after the White House and the heads of the biggest AI companies signed a voluntary, "morally binding" accord on internal controls and outside audits. Critics say voluntary is the problem. Supporters say it keeps the rules light while the government works out what it wants.
Why it matters for sellers: look at who's on the list. The FTC is the US agency that goes after fake reviews, misleading ads and consumer fraud. Having its chair on the AI task force suggests consumer harm, including AI-powered scams, is at least on the agenda. You don't sell in America, but the platforms you sell on are American, and their rules on AI content, ads and fake accounts tend to follow Washington's mood. Expect the next policy shift on Instagram or WhatsApp to arrive with this context behind it. For now, though: light rules, voluntary promises, a report in four months. Nothing in this changes your Monday.
Google shuts a bug-bounty door because AI junk flooded it
This one is small on paper and telling in practice. Google has suspended product-bug submissions to its Open Source Software Vulnerability Reward Program, the scheme that pays people for finding security holes in open-source projects. The reason: thousands of AI-written reports that described bugs that didn't exist, or that nobody could actually exploit. Google says it will share an update in early 2027.
It isn't alone. Linux maintainers have described being overwhelmed by AI-found vulnerability reports, and Intel has also paused its bounty program. When writing a report costs nothing, people send a lot of them, and the humans reading them run out of time for the real ones.
Why it matters for sellers: the same thing is coming for every inbox, including your customers'. When anyone can generate a hundred messages a minute, attention becomes the scarce thing. Two lessons:
- Don't become the spam. Blasting AI-written promos to your whole contact list is how you get muted. Our piece on WhatsApp broadcasts that sell instead of annoy makes the case for sending less.
- Use AI to answer, not to shout. The useful job for AI in a shop is replying fast and correctly to someone who already wrote to you. That's the opposite of a flood.
Reality check: shopping bots are still years away
Every week brings another "AI agent will shop for you" launch. The Register ran a sober counterweight. A product manager at Hedera AI Studio told the outlet that agentic commerce "is not a reality yet" and estimated roughly five years before it works properly. The reasons are practical: agents struggle with logins, page layouts and delivery math; payments need hard rules, not a model's best guess; and for expensive things, people still want to decide themselves.
The piece also pointed out that Amazon has blocked agents from Meta and Perplexity from its store. The big players aren't exactly rolling out the welcome mat.
Why it matters for sellers: relax a little. Your customer is not about to be replaced by a robot buyer. For the next few years, the person who buys from you is still a person, typing "price?" or «موجوده؟» in your DMs at 11pm. What decides the sale is how fast and how well that person gets an answer. That's a problem you can fix this week, not in 2031.
Quick takes
- ChatGPT Finances goes free in the US. Since October 2, Free and Go users in the US can connect bank accounts so ChatGPT can explain their spending. More people are about to ask an AI "can I afford this?" before they buy.
- WhatsApp tests "Restricted chat". A new privacy option in the iOS beta lets users tighten how others interact with a chat and keep specific chats off linked devices. Beta only for now.
- Reflection AI readies an open-weight model. The Nvidia-backed startup is preparing its first open model, pitched at Western companies wary of Chinese ones. More open models usually means cheaper AI tools down the line.
- Stability AI turns to music. Under Sean Parker, the image-generation company is reportedly pivoting toward AI music tools built with record labels, not against them.
The theme this week is simple: nobody upstream is promising to keep the bad actors out, and nobody is about to replace your human buyers either. What's left is the part you control, answering real customers quickly and carefully, on the channels they already use. That's the job Vardast was built for: it answers your DMs on Instagram, WhatsApp and Telegram around the clock, so the real buyers don't wait while you deal with the fake ones.