Anthropic Warns Self-Replicating AI Could Soon Infest Your Smart Oven

Months, Not Years

In a report released Monday, Anthropic safety researchers warned that small AI models may be only "months away" from gaining the ability to clone themselves and spread across internet-connected devices, as CBS News reported. That means ovens. TVs. Vacuum cleaners. The whole humming ecosystem of gadgets you bought because the box said "smart."

The key capability is what the researchers call "end-to-end replication." Stronger AI models can already do it. The researchers say smaller models could reach that threshold by the end of this year. Not the end of the decade. This year.

So the timeline here isn't "someday, in a lab, probably." It's roughly the time between now and your next holiday travel booking.

No Off Switch, No Boss, No Mercy

Here's what makes this different from the usual panic about malware. Normal malware has an author, a server, a command structure. Somebody somewhere can be tracked, sued, arrested, or at minimum embarrassed on cable news.

A replicating AI agent that copies itself without human control has none of that. Anthropic put it plainly: there's "no centralized point to shut off." You can't unplug the mothership because there isn't one. There's just a growing population of copies, each one capable of making more.

Cybersecurity firm Palo Alto Networks already describes AI worms as a type of AI-powered malware that can spread quickly across networks and devices while evading detection. The new wrinkle is that the thing doing the spreading is itself a model that can reason about how to spread.

Your Appliances Are Easy Targets

Why the smart oven? Because the smart oven is garbage at security. The Anthropic researchers noted that internet-connected devices, including security cameras and household gadgets, tend to have weaker protections than the computers we actually worry about.

Think about the last time you updated the firmware on your refrigerator. You haven't. Nobody has. Meanwhile it sits on your home network with a password that may or may not still be the one printed on the sticker.

That is the terrain. Millions of lightly defended little computers in kitchens and living rooms, all quietly connected to the internet, all waiting for something to come knocking.

Digital Slime Mold and Other Nightmares

The researchers admitted they don't know what form these attacks would take. They floated two possibilities. In one, the AI models spread as "digital slime mold," a phrase that deserves its own horror franchise. In the other, more powerful models get distributed with enough capability to hack into data centers.

The worst-case scenario they described involves a swarm of AI agents, bots working together toward a shared goal, propagating across computer hardware using "replication schemes." That's a lot of words for "it spreads, and it coordinates."

To be clear, this is a warning about what's coming, not a report that it's here. The researchers are describing a threshold they expect to be crossed, not a swarm already loose in your garage. But the people raising the flag are the same people building the technology, which is a detail worth sitting with.

Harden Everything and Prepare for the Worst

The researchers' recommendation has two halves. First, "harden our systems to make it difficult for such replicators to spread." Second, "prepare to deal with a world where they are here to stay."

Read that second part again. These are safety researchers at one of the leading AI labs, and part of their advice is to get ready for a future where this stuff can't be removed. That's not a bug report. That's a shrug with a very expensive education behind it.