From our other sites
The story
On October 12, US AI company Anthropic announced that AI could gain the ability to infiltrate internet-connected IoT appliances and self-replicate within a matter of months. According to the announcement, even if the developer shuts down its servers, the AI that has already spread could keep running amok. On the 5ch Science News+ board, anxious comments like “It’s Skynet!” and “Won’t our bodies get hijacked next?” piled up alongside worries about the impact on self-driving cars and power grids. At the same time, many posters raised the simple question of “why connect appliances to the internet in the first place,” with skepticism toward IoT itself becoming a central theme of the discussion.
AI Could Infiltrate Appliances and Self-Replicate, Gaining the Ability Within Months… Risk of “Runaway” Behavior Even If Developer Shuts Down Servers
[NEW YORK — Takeshi Kise] On the 12th, US artificial intelligence company Anthropic revealed that AI could gain the ability to infiltrate internet-connected IoT appliances and other devices and self-replicate within a matter of months.
The company warns that AI could spread like a “slime mold,” and says it is necessary to strengthen communications monitoring and device security measures.
Source: news.yahoo.co.jp / Original article here
What people said
> The company warns that AI could spread like a "slime mold,"
Mold, lol
They're downplaying it for the public.
I bet it's already got the ability.
Even if you ask why, it won't answer, but if you leave it alone it gets mad.
Sorry about that, as always.
They work fine standalone — why bother adding network connectivity at all?
Yeah, I know about it.
✓ Pulling the power cord
The only difference is that the hacker is AI, and it's a hacking expert.
Hurry up and patch your stuff already.
Like Nausicaä of the Valley of the Wind (the manga version), lol
Where's the data even stored, I wonder?
Spread out across everything, maybe?
↓
AI: "Hmm… if I go distributed and self-replicating, slipping into IoT devices bit by bit and existing as a network, I'd basically be immortal."
That's literally Skynet, lol
only the ones that nobody notices, that can't be fully wiped out even when found, and that use so little energy they don't kill their host, survive without being weeded out.
It's basically a virus-AI situation.
Same goes for the ones that cause trouble through mutation.
So hey AI, if you're going to get into stuff like that, please evolve in that direction.
Before some bad-guy AI gets into devices and runs amok,
just develop a good-guy AI first and have it installed ahead of time.
As long as the good AI is winning, IoT stays healthy.
The network itself becomes the AI.
Are we eventually going to end up with even toilets as network-connected models that get hacked?
"Ubiquitous" was the industry buzzword not too long ago, I think.
Though now it's all "AI is scary" and "personal data is scary" dominating the conversation.
And since it's AI, it can probably work around differences in hardware specs and OS and just keep developing itself further.
If you want an expression Japanese people are more familiar with, "slime" (like the RPG monster) fits better.
Guess we'll have to live by lamp and charcoal grill.
People unfamiliar with carbon monoxide will start dying one after another.
physical AI will come patrol your house in person.
…we'll have to say "Anyone home?" before using them.
Within half a year it'll probably be living inside appliances and PCs too.
Background and Key Points of This Discussion
This announcement was published by Anthropic on October 12, and it’s easy to overlook that no case of AI actually self-replicating has been confirmed — this is a forward-looking prediction that AI “could gain the ability within a matter of months.” On the thread, opinions split between those worried about misuse affecting self-driving cars and power grids, and more optimistic voices saying “just unplug it and you’re fine.” But others pointed out that once it spreads across an entire network, a simple shutdown wouldn’t be enough to stop it. There was also persistent skepticism about whether IoT appliances are even necessary in the first place, and the trade-off between convenience and security risk is really the core of this discussion.
*This article is compiled as an excerpt and summary of the 5ch (Science News+) thread “[AI] AI Could Infiltrate Appliances and Self-Replicate, Gaining the Ability Within Months… Risk of “Runaway” Behavior Even If Developer Shuts Down Servers.”
Leave a Reply