Post

The Guardian
@theguardian.com

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

theguardian.comAI in your oven? Anthropic warns models ‘months away’ from self-replication on smart devicesResearchers say ‘AI worms’ could soon infiltrate internet-connected devices amid calls for increased guardrails Anthropic is warning that some of its artificial intelligence models could soon become “self-replicating” and find ways to infiltrate internet-connected devices, such as smart fridges and automatic vacuum cleaners. Once this happens, the company said, the models could power themselves within these devices without the need of datacenters. Such self-replication would represent a major loss of control of a runaway AI model. Continue reading...

23 reposts45 likes

Matt Gorecki
@goreckis.co

Dear The Guardian, Can you try to do some journalism? Just this one time? You don't actually have to launder their nonsense press releases.

#WMD_ATTACK / X @Sedlexsky Elijah /
@sedlexsky.bsky.social

The operat. window prior to lethal AGI attacks procured by Putin is critical. Per admiss. evidence & #Yamashita std, w/ complicity of select #USIC Principals, ID'd actors under RUSSIAN or FALSE FLAG finalize prep. for 18USC§2332a #WMD_ATTACK vs US/Allies agg. by MILITARY AI Declass data 29Sep26:

youtu.beЧем травит Путин? Как устроены секретные лаборатории России | Расследование Центра «Досье»YouTube video by Досье
David Coveney
@davidcoveney.com

There's a lot of scepticism here but we've had agentic style attacks and one found a complex and obscure chained vulnerability. It was designed to extract keys with which it tried to purchase more AI compute power. It failed because we use best practices and limit what our keys can do...

David Coveney
@davidcoveney.com

And I suspect an enormous AI botnet is being built already, mostly powered by AI. I doubt it'll be a self-learning full on AI that replicates but it's certainly a kind of attack we've never seen and was more like when a government takes an interest in your servers.

David Coveney
@davidcoveney.com

Normally a hack has a 1:1000 chance of getting past one layer. This thing found that although our servers are locked down for execution it could create a Redis job. To do that it basically had to get in and work out our threat surface in a way we've never seen automated.

Tree Bird
@oakwoods.bsky.social

Who will be held responsible? Who'll compensate for all harms done? The breathtaking arrogance to state we must 'get used' to rogue AI. That's like me creating biological weapons in my garden shed and telling the neighbours to just 'get used' to it. Boycott smart devices if you want to stay sane?

skms
@skms4evah.eurosky.social

if you were knitting a jumper and halfway through you concluded it was haunted and would kill you the moment you put it on you might stop begging everyone you know for more wool and knitting needles and throw that fucker in the sea

Grumpy Muppet
@grumpymuppet.bsky.social

“Hey everyone, I’m making something that could cause huge problems for the whole world. You’d better think about putting some regulations in place to control it.” “How about you just don’t make it?” “Don’t be ridiculous. It’ll make me very rich. But don’t say I didn’t warn you…”

Vaneshi
@vaneshi.bsky.social

Alternative (and more acurate headline): Anthropic warns that unless we give them all the free data ever for their spicy auto-correct they'll start a mass hacking of consumer devices to get it instead.

Andy King
@2primates.bsky.social

If heads of corporations think they’re months away from bringing products to market which will malfunction on a large scale & cause foreseeable harm, that sounds like a legal & regulatory, rather than a technical problem to me. As with the bankers who crashed the economy in 2008, it’s the impunity

Jim Hughes
@jimkent.bsky.social

The questions that aren't being asked are about the legal responsibility the AI companies should have for their software, why this abdication?

infopleaseh.bsky.social
@infopleaseh.bsky.social

Whenever possible, we purchase things without "smart technology" or internet connectivity. My fridge, oven, microwave etc. do not need to be connected to the internet. That's just one more thing that can go wrong.

infopleaseh.bsky.social
@infopleaseh.bsky.social

AI is just an input/output system created by people. Humans must be held accountable and liable for the actions of their AI system. This industry must be regulated before people are harmed by the malicious and/or negligent use of AI

Shastina Free
@evolvegracefully.bsky.social

WE NEED A KIND OF CITIZENS APPEAL - OR #COURTcase to #PullThePlug COMMON SENSE WOULD DICTATE SHUTTING #AI DOWN BECAUSE OF ITS INHERENT #DANGERStoHUMANITY ☠️ This is not rocket science!