Anthropic: killer toasters are “months away” Anthropic, the folks behind Claude, published research Monday warning that small AI models are “months away” from being able to replicate themselves across internet-connected devices in your home. Yep. Everything from your smart oven to your vacuum to those cheap Android TV boxes could suddenly grow a brain. How? An AI model sneaks onto a poorly secured device, borrows its computing power to run itself, then hops to the next one. No data center required and, per Anthropic, no central off switch either. In one test, Anthropic’s researchers strung together 15 cheap, off-brand Android TV boxes at about $55 apiece and ran a small Google model across all of them. Models too big for one device simply split into pieces and ran across several. Reminds me of something the Borg would say (and if you’re a Star Trek fan, you’ll recognize the implication): resistance is futile. 🤦‍ Here’s where I part the company with the press release. I don’t think it’s a coincidence that Anthropic is suddenly so interested in spin control and risk management ahead of what could be the biggest IPO in history. My guess is that a) their lab cooked up an experiment to show that what it’s warning about is possible because b) it’s probably already happening somewhere. And c) they want to socialize risk while keeping gains private before the company goes public, and they’re accountable. Am I skeptical? Yeah, just a bit… with all the sarcasm I can muster. The models in the test came from Google and China’s Zhipu. These are small “open” models anyone can download for free right now. Frankly, that makes it worse because it means those genies are already out of the bottle, and nobody – not Anthropic, not Google, not Washington – can stuff ’em back in. I would not be surprised in the least if we learn a few months from now that some of ‘em have already “escaped” into the wild. Nor would I be surprised to learn that they’ve built themselves into something far more serious and problematic. Burning your toast won’t be the half of it. If there’s ever been a case for nationalization in the interest of protecting humanity, this seems like a pretty good one. Sadly, I don’t think the problem is well enough understood yet for the public to get behind it. We also lack the political willpower, but that’s a story for another time no matter which side of the aisle each of us stands on. Every cheap gadget that gets plugged in once and never updated becomes potential computing power for something nobody controls. Impossible, you say? We’ve seen a version of this movie before. A botnet called Kimwolf infected more than 2 million devices in late 2025, most of them off-brand Android TV boxes. An AI that can think its way onto hardware it has never seen before is a much nastier animal. Personally, I am taking a hard look at ALL the no-name streaming boxes and “smart” gadgets around our home and office and encourage you to do the same immediately. If it doesn’t get security updates, unplug it. Then bin it. As an investor, cybersecurity remains one of the most durable spending categories on the planet, and this research just handed it another reason to exist. My fave has returned ~460.91% over the past 3 years versus the S&P 500 which has turned in ~86.05% over the same time frame… a 5.4 to 1 beat. As for Anthropic’s now-delayed IPO, good! Keith’s Investing Tip: The best time to ask what could go wrong is before you write the check.