Anthropic says AI models could soon copy themselves across your devices — with or without you

It’s spooky season, and one major tech company has made a discovery that could rival some dystopian sci-fi movies.

Anthropic warns that small artificial intelligence models are just months away from becoming capable of replicating themselves across internet-connected devices in your home. The company says small AI models could potentially get into household items that run on what’s known as the Internet of Things (IoT) — like smart refrigerators or Android TV boxes — and not only run themselves, but also spread to other IoT devices, without any help from a data center.

And it gets more troublesome from there. Anthropic says it would be “challenging to shut down, since there’s no centralized point to shut off.”

The jargon

Understanding how the models replicate can get pretty confusing if you’re not a tech whiz, so here are the terms you’ll need to know.

• AI model: Microsoft defines an AI model as “the engine inside an artificial intelligence system that learns from data to perform tasks. It combines algorithms, training data, and learned parameters to transform raw inputs into outputs like recognizing speech, predicting equipment failures, or generating new product designs. AI models work at the intersection of artificial intelligence and machine learning, where algorithms continually learn from data to deliver more accurate predictions and better responses over time.”
• Small AI or Small Language AI model: Often shortened to SLM, these AI models run on millions to a few billion parameters and are designed to run fast on edge devices, phones, or local computers, according to IBM.
• Edge devices: Hardware components that sit at the boundary or “edge” of a network, processing, filtering, and storing data locally near where it is generated rather than sending every raw piece of data to a centralized cloud or data center, according to Hewlett Packard. These include things like phones, smart thermostats and autonomous vehicles.
• Internet of things: More commonly known as IoT, this is a network of physical objects embedded with sensors, software, and network connectivity that allow them to collect, share, and act on data in real time without human intervention, Microsoft says. Virtually anything in your home described as “smart” applies here. 
• Application Programming Interface (API) key: A unique identifier used to authenticate software and systems attempting to access other software or systems, per IBM.

What could happen

In a blog post published Monday, Anthropic said there isn’t yet a “fully rogue replicator.” The company’s researchers found that when there is, small AI models may be slow to replicate, possibly even taking days to do so. But they will have the capacity to figure out unfamiliar devices on their own — distinguishing them from non-AI malware.

“Ever-improving AI capabilities make a longstanding concern increasingly realistic: autonomous replication and adaptation (ARA), which involves a model deploying and propagating itself without human control,” Anthropic said.

Anthropic also said these autonomous AI systems are likely to rely on widely ⁠available, relatively insecure computing resources to replicate.

“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 company advised.

In one test, the company ran one of Google’s SLMs across a network of 15 inexpensive off-brand Android TV boxes, which often have weak security. Researchers found that even a model too large to fit on one device could “split into pieces, or shards” and run across multiple devices.

“In practice, an AI worm wouldn’t necessarily consist of just one model or strategy,” Anthropic’s blog post read. “A swarm of different models deployed on different types of hardware using a variety of replication schemes might represent a more realistic ‘worst case scenario’ attack.”

Anthropic’s researchers are still “uncertain exactly what form autonomous AI replication will take.”

Bigger concerns

The new warning comes as leaders at Anthropic and other tech companies, like its rival OpenAI, have repeatedly warned about the potential risks of AI as the technology advances rapidly. Some are even calling on the government to impose stricter guardrails to slow that growth.

Now that Anthropic is raising concerns that once these SLMs completely gain the ability to replicate autonomously, that ability could evolve even further.

“The rise of slow replicators would also be a bellwether for more aggressive, faster ones that can breach more hardened machines or be effective API key thieves,” Anthropic said.

Round out your reading

• He built his own Flock-style camera to track police. The police didn’t like it.
• Trump’s apparent joke about letting Iran bomb LA and San Diego catches bipartisan condemnation.
• Grandparents are fighting for access to their grandchildren. Here’s what the law says.
• When the ‘victim’ is made with AI, child pornography cases get complicated.
• Fibermaxxing is trending. Here’s what the science says.