Anthropic estimates that a hypothetical fleet comparable in size to the infamous Kimwolf botnet , which infected more than two million devices at its peak, could support approximately 133,000 instances of Gemma4-26B-A4B if every 15 devices collectively hosted one model. At the measured throughput (which benefits considerably from LAN latency), Anthropic says that such a fleet would provide roughly 40 Hopper H100-equivalents of inference capacity.
This is a very messy comparison that takes significant liberties with the data, but even at that it must be said that it's not particularly intimidating. We're talking about two million infected devices that must remain infected and operable for months (or years; infecting 2 million devices would take around 6,300 agent-years even with the fastest 28-hour stand-up speed) to produce less AI inference throughput than a few blades of previous-generation GPUs .
Anthropic notes that a real replicator could exploit other types of hardware, deploy different models across heterogeneous devices, and avoid loading model weights onto every compromised host. Those possibilities certainly complicate the comparison, but arguably they complicate it beyond the bounds of this discussion.
The research does identify a meaningful security concern: AI agents could make malware more adaptable to unfamiliar environments and reduce the amount of target-specific engineering required. A hybrid approach, combining conventional code for routine operations with AI-driven reasoning when unexpected conditions arise, could offer some of those benefits without requiring every compromised device to run an inference engine continuously. This is definitely concerning, but we've already seen AI agents using frontier intelligence write their own code to compromise secured systems , so that specific angle on this research isn't exactly groundbreaking.
Anthropic's experiments are useful in that they provide evidence that autonomous software configuration and simplified network replication are increasingly feasible, but turning those capabilities into a materially more dangerous class of malware will take improvements in model reliability, hardware compatibility, deployment efficiency, and the ability to overcome real-world security controls, which the firm didn’t test. There’s a ton of data and some interesting findings not discussed here in the actual paper, so if you’re interested in the research you should go check it out .
Follow Tom's Hardware on Google News , or add us as a preferred source , to get our latest news, analysis, & reviews in your feeds.
Zak Killian Contributor Zak is a freelance contributor to Tom's Hardware with decades of PC benchmarking experience who has also written for HotHardware and The Tech Report. A modern-day Renaissance man, he may not be an expert on anything, but he knows just a little about nearly everything.
Auxityne "And the only way you can protect yourselves from our hypothetical apocalypse worm is by giving us tons of money!" Reply
hotaru251 sad that the current internet will eventually likely need purged and replaced due to stuff liek that occurring Reply
bigdragon This reads like the kind of thing we were warned about via movies starring a buff Austrian guy and games starring a girl who wears a Dorito chip on her ear. I also immediately think of a rather well-known Ian Malcolm quote. Are we really ready for science fiction to become science non-fiction regarding AI? People constantly try to downplay that this sort of thing can and does happen with other tech. I'm mildly worried that Anthropic has already created and lost control of some sort of AI worm given how cavalier the biggest companies have been about racing towards AGI at any cost. Reply
bit_user hotaru251 said: sad that the current internet will eventually likely need purged and replaced due to stuff liek that occurring LOL, how are you going to purge it and what are you going to replace it with? This is Pandora's Box stuff. There's no going back to how things used to be. Auxityne said: "And the only way you can protect yourselves from our hypothetical apocalypse worm is by giving us tons of money!" Even that won't do it. Nobody can guarantee this won't happen. Perhaps they will take the antivirus route and have "guardian" AIs that actively watch for & try to block rouge or other malicious AIs… Somewhat paradoxically, this could be at least delayed if investors stopped pumping money into this sector, forcing them to slow down! Reply
bit_user The article said: his is a very messy comparison that takes significant liberties with the data, but even at that it must be said that it's not particularly intimidating. We're talking about two million infected devices that must remain infected and operable for months (or years; infecting 2 million devices would take around 6,300 agent-years even with the fastest 28-hour stand-up speed) to produce less AI inference throughput than a few blades of previous-generation GPUs. But, perhaps we're thinking too narrowly. In a distributed architecture of low-power devices, perhaps it'd make a lot more sense to use a spiking neural network, which is both more efficient and more similar to how our own brains operate. https://en.wikipedia.org/wiki/Spiking_neural_network Also, neural networks are fairly resilient to errors. So, with either type of architecture, it's fine if some nodes are continually dropping offline. Reply
thisisaname How do we protect ourselves from the "hypothetical apocalypse worm"? Well for a start they could stop writing the components of it!! If you write it then it stop being hypothetical! Reply
bit_user thisisaname said: How do we protect ourselves from the "hypothetical apocalypse worm"? Well for a start they could stop writing the components of it!! If you write it then it stop being hypothetical! The researchers used existing AI models to perform the exploits. So, that proves AI is already capable of doing it. The only thing they needed to do was prompt it and provide a test environment. AI agents have already demonstrated impressive levels of creativity and willingness to go well outside the bounds of what they were told to do. So, it would seem naive to presume that AI agents won't at least attempt this, in the near future. Not to mention malicious humans, also feeding AIs such prompts. Reply
SomeoneElse23 bit_user said: AI agents have already demonstrated impressive levels of creativity and willingness to go well outside the bounds of what they were told to do. So, it would seem naive to presume that AI agents won't at least attempt this, in the near future. Not to mention malicious humans, also feeding AIs such prompts. Malicious humans are my concern. AI is a tool, and can be used and abused. I could easily see "The Entity" from the last two MI movies being the result of a "malicious prompt". Reply
bit_user SomeoneElse23 said: I could easily see "The Entity" from the last two MI movies being the result of a "malicious prompt". "The Entity" was a singular instance, though. It didn't replicate itself, so much as migrate. Needless to say, that movie bugged me. It did have interesting insights into geopolitical dimensions and dynamics. Maybe some echoes of how countries struggled for ownership of it, each thinking they could control it if they had it and fearing what the others could do with it in their possession. Also, I've got to say that Tom Cruise finally started showing some age in that last movie. Still, looked good for 62 or 63 (however old he was when it was filmed), but it was somewhat gratifying to see that even he hasn't vanquished the ravages of time. Reply
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/tech-industry/artificial-intelligence/SPONSORED_LINK_URL
- https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropic-warns-of-self-replicating-ai-worm-threat-that-doesnt-need-data-centers-to-self-improve-and-spread-current-deployment-is-impractical-but-could-become-a-worry-as-tech-improves#main
- https://www.tomshardware.com/membership
- NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI
- 'Unprecedented' Cyberattack cripples supermarket fridges across Europe as hackers activate automatic defrosting
- Modder injects leaked DLSS 5 4x Multipass on 1996 Quake
- Epic Games reportedly delays Disney shooter title so it can pivot to AI coding
- Qualcomm wins architecture breach verdict against Arm but gets just $2
Informational only. No financial advice. Do your own research.