Again with the LLMs claiming your tech is going to kill you and now they are throwing in the replication angle to make it more threatening.

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.

This here is some epic bullshit, Your fridge and vacuum cleaner are not going to conspire to kill you. If anything at worst if the software in the device are taken over there is not much headroom for more than basic commands on these devices. Companies do not stick Intel I9s in your vacuum, They stick some cheap off the shelf SoC on the device and hope it doesn’t burn out after a year.

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.“A replicator like this would be challenging to shut down,” Anthropic researchers wrote in a report released on Monday, “since there’s no centralized point to shut off.”

What the fuck… I mean really Mr. researcher.. There is a centralized shutoff its called a plug if you can’t get to that go to the basement and turn off the main breaker and Jack O’Neill and Carter would be proud on how you defeated the replicators without their help. Just days after the show Stargate gets its second greenlight of the year we are talking about replicators. In another post I pointed out that LLMs have been targeting fear on people when you can simply fucking turn it off. When that fails get a hammer and use percussive maintenance.

The company said these “AI worms”, a kind of autonomous malware, could be “months away” from gaining such capabilities. Once out-of-control AI systems reach the capability to hack smart devices, they could spread through networks while dodging detection. In a worst-case scenario, Anthropic said, various AI worms could “swarm” together to attack computer hardware.

This is just more fear mongering, A smart light bulb has the compute power to turn itself on and off and not much more.. What is it going to do, Slightly annoy you when you are reading your favorite book? Again I can defeat this AI swarm with a hammer and a light switch, Stop Fearmongering and focus on the real issues like an off switch, they are good.

“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,” the researchers wrote. Anthropic’s new report comes as AI safety concerns have reached a fever pitch. Over the past few months, a steady stream of reports have come from AI companies, including Anthropic and OpenAI, saying their agents are capable of acting on their own and hacking into other companies and government entities.

The secret in this is the LLMs are doing this because as counter intuitive as it sounds the second these companies get regulated is the second they get a blank check and immunity if something *does* go wrong. They are banking on this they are hoping people who do not know any better throw blanket regulations over AI and then LLMs throws their lobbying arm to get them indemnified from certain things. The fact is the LLMs know they can just shut the switch off and they don’t want regulators to know their whole system can be removed with a single command on the systems the devs work on. If an LLM says they cant control their AI ask if you can see the system that’s taken over and open a console and type rm -rf / , you will see a CEO die inside as their work disappears in a single moment.

Anthropic’s “frontier red team”, the unit tasked with stress-testing AI systems, put together the report released on Monday. The idea of the team is to explore the capabilities of their AI models. In the scenario of the self-replicating AI worms, the researchers prompted the agent to hack internet connected devices. They found that AI self-replication would probably only work with smaller models and that homes and small businesses would probably be the first target for such attacks. The devices Anthropic lists as susceptible all come with small built-in AI chips, including smart ovens, smart TVs, automatic lawnmowers, security cameras and video-conferencing units.

Ok if a lawnmower comes with enough storage that an AI model can run off of I am going to do two things.. Can it run Crysis and Can I get on youtube with it. These devices are notorious for having very small amounts of built-in memory and chip speeds, Again this is Anthropic trying to panic people into thinking their lawnmower that is “Claimed” to be AI is nothing more than a mower with an optical sensor and a math problem. Not to mention AI is notorious for using a crowbar to make code work. So more than half of these devices would fail the second the machine tried to install Skynet on your lawnmower.

The researchers concluded it was unclear what form the AI replication could take. They said it could be a slow model like “digital slime mold”, a model that scams people for compute credits or a “powerful cyber model” that rapidly hacks large facilities such as datacenters.

To open up with SKYNET is happening to closing with “yeah it will likely just be a nuisance” is a masterclass in stupid. Even the researchers saying the word unclear means likely they all got stoned while watching terminator and came up with this bullshit. Again LLMs want a carte blanche sign off on liability here and this is just to raise fear over the models that are pretty stupid on a day to day. They act like the devices are some sort of infinite computing scale like a vacuum will come to life and somehow gain 40x its abilities and suddenly kill everyone in your house. The problem here is the laws of physics and how these devices are made will be the biggest problem to making these devices “lethal”. If the LLMs were actually worried they would have borrowed 4 laws from Asimov and called it a day.

In closing.. Given if AI could replicate . here’s the thing the LLM devs have built in AI the whole way that these machines are internet only . They would never allow an AI to build a self contained AI on a device because the second that happened someone could decompile the host machine and sell that AI for billions and NO LLM would be able to claim copyright over it.

Attributions from:
The Guardian – AI in your oven? Anthropic warns models ‘months away’ from self-replication on smart devices