Anthropic warns AI-Powered 'worms' could infect your smart devices
We have enough to worry about when it comes to artificial intelligence weaving its way into our lives, but now, one of the biggest AI companies around has issued a warning about AI-powered 'worms' infecting our everyday tech.
Are we really in danger of our homes becoming The Simpsons’ villainous Ultrahouse 3000? It's not just the idea that your Amazon Alexa could be listening to you, with more recent fears that everything from our robot vacuums to air fryers are spying on us.
Hackers have previously shown how easy it is to gain invaluable information about our personal lives using something as innocuous as a smart lightbulb, but according to Anthropic, these worms are another thing we need to worry about as AI continues to evolve.
What has Anthropic said about AI worms?
Anthropic has issued a PSA that AI could be just months away from being able to replicate itself on commercial internet-of-things (IoT) devices and behave like the kind of worm viruses you might find on your home computer. In a lengthy blog post, Anthropic mentioned the likes of smart ovens and off-brand Android TV boxes, reiterating: "A replicator like this would be challenging to shut down, since there’s no centralized point to shut off—though it may be slow (with a doubling time of days)."
Advert
Unlike non-AI malware, AI worms could also figure out unfamiliar devices on their own.
Worms are programs that copy themselves from one computer to another without the need for human input, with previous work showing that AI agents can act in a similar way as they scan a network, execute an exploit, and copy model weights to target machines across a GPU cluster.
Data center GPUs are obviously a high-profile target, but according to Anthropic, the cheap and poorly defended range of IoT devices is where we should be focusing our attention.
Anthropic's researchers based calculations on how far these mythological worms could spread on Kimwolf, a botnet that infected over two million devices at its peak and largely infiltrated off-brand Android TV boxes.
Like signs of life were seen in coding agents in 2024, Anthropic says we're seeing the same for AI worms in 2026. Although there isn't yet a rogue replicator, agents are supposedly becoming increasingly capable of completing steps in the replication chain.
Even though AI replication could be a slow distribution that's dubbed 'digital slime mold', it could also be a major attack that breaches data centers at an alarming rate.
What devices are most at risk of AI worms?
Jordan Rae Kelly, senior managing director and head of cybersecurity at FTI Consulting, spoke to CBS News with her thoughts. Discussing replicators, Kelly said that we're moving into an era where we don't need human interference to create bad actors: "You could be looking at potentially destructive acts happening in environments that weren't predicted."
She reminded us that the average American household likely has 'dozens' of IoT devices connected to their network, often without even realizing: "These are devices that we like the connectivity of them because we can turn on our dishwasher while we're away, but this could become a breeding ground for bad activity.”
There's also a somewhat alarming conclusion as Anthropic added: "While it’s true that IoT devices have been vulnerable for a long time, lightweight agents may be much more willing than human attackers to put in the effort to compromise large fractions of the world’s IoT compute."
Not just our smart TVs potentially putting us at risk, she also points to refrigerators and Ring doorbell cameras while reminding us that we're moving at a speed and a scale that wasn't anticipated before. Kelly says we need to show we have our defenses moving at the same rate, praising the research for coming out now but clarifying: "It's not as simple as just turning off the capability and the capacity because it exists."
As for what we can do, Anthropic reiterates: "We should aim to harden our systems to make it difficult for such replicators to spread, but also prepare to deal with a world where they are here to stay."
So, next time you think about putting your trust in your AI kettle, just remember it could be a silent assassin lurking on your kitchen countertop.