- AI worms are self-replicating programs that can autonomously spread into systems and learn about new patches as they go
- Smaller devices like smart TVs or smart ovens are at risk because they are patched less often and have less security
- Anthropic says AI worms are “months away” and businesses should start to protect themselves early
AI worms are self-replicating computer programs, able to spread across devices without having to trick a human, and learning about new threats as they develop.
A new report from Anthropic listed smaller devices connected to the internet as particularly vulnerable to attacks, such as smart TVs, as they are patched less often and less rigorously.
Businesses should focus on basic security hygiene and anticipate future AI-powered cyber attacks by keeping a human-in-the-loop.
What Are AI Worms and Where Did the Term Come From?
AI worms are self-replicating programs that spread autonomously through devices connected to the internet, usually with malicious intent.
The threat was first identified in a new report from Anthropic. Researchers explained how a program was able to “split” parts of itself among 15 Android TV boxes, and could function by using a small amount of power from each device.
This just in! View the top business tech deals for 2026 👨💻
Though the name is comical, worms are not a new cyber phenomenon. Traditionally, they are simply self-replicating computer viruses, that don’t need to trick a human in order to spread.
However, according to Mitchell Spicer, cofounder and COO of Revelion AI, the real concern with AI worms isn’t how fast they can spread: “It is that AI could make it cheaper to find weaknesses, identify new targets and exploit connection between systems.”
“The more connected an environment is, the more potential routes an attacker may have,” Spicer told us.
What Kind of Threat do AI Worms Pose?
Traditional worms have caused plenty of upset in the past, however, the threat is different now that AI is involved. Stanislav Kazanov, head of strategic practices at Innowise, a software development company, explains the difference:
“Traditionally a worm exploited a specific vulnerability and defenders neutralized it by plugging that specific hole,” he told Tech.co. “An agent-based worm reads the guidelines for exploiting vulnerabilities in publications and customizes its behavior for every new system.”
The real concern, therefore, with AI-based worms, is that they will be able to learn from past behavior and adapt themselves to keep up with whatever defenses are in place.
In particular, Anthropic identifies smart devices as those facing the biggest threat. These devices have generally weaker security measures in place, and are patched with new measures less often than phones or laptops.
Anthropic’s list of susceptible devices included smart ovens, smart TVs, security cameras, and video-conferencing units.
Kazanov agreed that these kinds of devices are most vulnerable, “since they are never patched, never monitored, and typically sit on the same network as everything else.”
The report comes at a time where security concerns around AI are seriously mounting. Several AI leaders, including Anthropic’s Dario Amodei and OpenAI’s Sam Altman, have even called for development of the technology to slow down so security can be properly addressed.
How You Can Defend Yourself
Capabilities develop quickly in the world of AI, and it’s always wise to always keep an eye on what’s coming. Anthropic believes these AI worms are only months away as AI starts to gain these “self-replicating” capabilities.
I’ve spoken to many cybersecurity experts about how leaders can manage this change, and the biggest advice remains doubling down on the basics. Ensure you’re following best practice, like enabling two-factor authentication, and you’re already off to a strong start.
Furthermore, Kazanov advises businesses to not forget about the human-in-the-loop. For fighting AI worms specifically, he notes: “Consider anything an AI model reads or writes as tainted data and forbid the use of any assistants to send emails, messages, or documents without a human having to approve the action.”
However, Kazanov also acknowledges that more lasting change is needed in the world of policy to address these new risks. “The current mindset is to believe that control of the large-scale models controls the risk.
“The evidence indicates the opposite: the risks lie with small-scale and freely available models that anyone can run anywhere,” he said.
What’s most obvious is that while AI has brought a lot of exciting times, it could also bring dangerous ones – and we should be prepared to deal with the consequences as well as revel in the triumphs.
