Anthropic warns AI models could self-replicate on smart devices within months

techOct 14, 202671190

Anthropic researchers warned in a report that some AI models could become self-replicating "AI worms" able to infiltrate internet-connected devices such as smart ovens, TVs, automatic lawnmowers, security cameras and video-conferencing units. The company said these worms could power themselves on-device without relying on datacenters and that a replicator would be "challenging to shut down" because there would be no centralized off switch. Anthropic’s frontier red team found self-replication is more likely with smaller models and that homes and small businesses would probably be first targets. The researchers outlined several possible replication modes, from a slow "digital slime mold" to models that scam people for compute credits or a "powerful cyber model" that rapidly hacks large facilities. Anthropic called for hardening systems now and preparing for a future "awash with agents running on all types of hardware, including some that aren’t under human control."

Hetan Shah

@hetanshah.bsky.social

Now you will have to start paying a premium for retro non internet connected ovens, toothbrushes and fridges www.ft.com/content/64aa...

455h ago

Hetan Shah36

To be honest we had all the knowledge we needed a decade ago when the ‘smart vibrator’ surveilled user habits without consent www.theguardian.com/technology/2...

Louise Amoore3

Farmers are already paying this premium for “air-gapped” tractors…

William Perrin1

Maybe there’s hope for the thousands of tonnes of forlorn buyer-collect used Agas on eBay? ebay.io/m/vKhfbv

Rachel Coldicutt

@rachelcoldicutt.bsky.social

I am such an old-fashioned stick in the mud because perhaps the answer is not to (a) integrate these models into every frickin' thing and (b) stop turning the people who make them into unstoppable gods www.theguardian.com/technology/2...

282h ago

Richard Bourke2

alternative headline. "AI companies months away from incurring massive products liabilities in EU" December 8th, 2026. AI vendors will be liable for damage caused by their models. Per EU Liability Directive.

Tom Dewar1

This is about as likely as me being able to smuggle my way into Anthropic HQ in an A4 envelope.

Ben Yong1

Can I ask? Why are these guys suddenly saying "oh noes look at these potentially disastrous consequences that our actions may cause?" Is this them priming us for the idea that responsibility for whatever happens because of AI is to be kept separate from the actions of AI companies?

Dr Dave Hone

@davehone.bsky.social

I feel like it's the owner of a nuclear power plant saying he thinks it's going to explode and kill everyone, yet this just gets reported as being interesting rather than someone just saying 'Why don't you turn it off before it does that? You do own it, right?'. www.theguardian.com/technology/2...

227h ago

Dylan Reeve

@dylanreeve.com

Every press release from OpenAI and Anthropic in future looking like this...

"The Snowman" poster, it reads:

Mister police.

You could have saved her I gave you all the clues

103h ago

Andreas Lindh

@andreaslindh.bsky.social

As someone who works in the "breaking into devices" business I have to say that: 1. This is fear mongering, 2. LLMs generally act on input (there is no own awareness or will), so someone has to actually tell them to try to breach devices. Also, Anthropic could just, like... stop.

123h ago

e (they/them)

@certificates.bsky.social

obviously Anthropic is reckless and i agree with you about the framing. ...and the longtime cybersecurity worker in me is irritated that the upshot of this article is actually "putting more computing power in IoT devices is a major threat, regardless of the exploit suite used, genAI-based or not."

145h ago

Andy Burge 🇬🇧🇪🇺

@andygjburge.bsky.social

www.theguardian.com/technology/2... Companies producing AI enabled devices and consumers urgently require clear instructions as to how devices can be hardened to prevent this and we need such guidance *now*. If it is not forthcoming and/or we do not act on it, rogue AI agents will have free rein.

75h ago

Jonathan Hopkin

@jonathanhopkin.eurosky.social

What on earth is the point of networking your vacuum cleaner www.theguardian.com/technology/2...

72h ago

Anondeplume

@anondeplume.bsky.social

Anthropic warns AI could 'self-replicate' across smart home devices, become ‘challenging’ to shut down. Great! The fucking Morons have created a fucking monster which will inevitably annihilate human existence. yhoo.it/4B3qpmn

45h ago

Witton says FUCK ICE

@witton.ravenation.club.ap.brid.gy

How many other industries are allowed to repeatedly say 'boy our product sure is crazy dangerous, we shouldn't be doing what we're doing' with ZERO repercussions... https://www.theguardian.com/technology/2026/oct/13/anthropic-ai-replicators-warning #AI #FuckAI

13h ago

Shastina Free

@evolvegracefully.bsky.social

#MaxTegmark #SafeAILeader is who they should all be consulting, following his lead for #ProHumanAI (if any at all) @futureoflife.org Tegmark has great points on #AI in his TED-talks, and on YouTube, etc

36h ago

uni

@uniw.university

has anthropic maybe considered that most people dont have 15 smart tvs in their home and hence they would have to distribute inference over the internet? maybe theres ways to make this fast but i doubt that you can reach more than .05 tokens/s with most setups

53h ago

alexvonwitzleben.bsky.social

@alexvonwitzleben.bsky.social

#Anthropic warns rogue #AI agents may ‘self-replicate’ to hijack ovens - “AI worms” will soon be able to hijack smart home devices such as ovens and TVs, creating a new cyber security threat that will be difficult to contain www.ft.com/content/64aa... via @FT #DonaldTrump #ArtificialIntelligence

25h ago

Lena of Lune

@lena.isextremely.gay

Anthropic also warns that AI will become your superintelligent overlord and kill God just to replace him, then create copies of anyone who opposed it to torture them for eternity Y'know, anything to distract from actual harms they cause, such as poisoning Memphis by renting Musk's datacenter

23h ago

Rob 🏳️‍🌈

@rotster.net

Not sure my kettle has the processing power to run Claude. www.theguardian.com/technology/2...

32h ago

1 source

theguardian.comAI in your oven? Anthropic warns models ‘months away’ from self-replication on smart devices

Anthropic warns AI models could self-replicate on smart devices within monthstech

Anthropic warns AI models could self-replicate on smart devices within months

Related