Anthropic has warned that small artificial intelligence models may soon gain the ability to self-replicate and spread across internet-connected devices; a capability that could expose televisions, security cameras and other smart home equipment to new cyber threats.
According to ITNA, Anthropic, the developer of the Claude chatbot, has warned about the emergence of a new generation of cyber threats known as “AI worms”; programs that can replicate themselves and spread among internet-connected devices without direct human intervention.
Based on a report published by the company’s safety researchers on Monday, October 12 (Mehr 20), small AI models may be only months away from attaining the ability to independently replicate and propagate across various devices.
This warning is raised especially regarding smart home equipment; devices such as smart TVs, internet-connected ovens and robotic vacuum cleaners that, due to their constant network connection, may become targets for such threats.
Replication without human intervention; a new challenge for cybersecurity
One of Anthropic’s chief concerns is the difficulty of containing AI programs that can replicate without human oversight.
According to the company’s researchers, once such programs spread across devices and networks, there will not necessarily be a central point to stop their activity. This feature could make detecting, controlling and removing them considerably more difficult.
Anthropic has stated that the most powerful AI models currently have the ability to perform a process called “end-to-end full replication”; a capability that allows a model to carry out the steps necessary to create another version of itself.
The researchers believe smaller models may also reach this level of capability by the end of 2026.
This development is worrisome because smaller models can operate with more limited processing resources and therefore could be run on a much wider range of devices.
Why are smart home devices at greater risk?
Cybersecurity firm Palo Alto Networks explains that AI-equipped worms are a type of malware that can rapidly spread across networks and devices and employ methods to evade detection.
Anthropic’s researchers have also warned that smart home equipment and internet-connected devices, including CCTV cameras, may be vulnerable to these threats.
The main reason for this concern is the relatively weaker security level of many internet-connected devices; an issue that can provide an opportunity for intrusion and the spread of malicious programs.
In such circumstances, an AI program capable of independent replication could move from one device to another and gradually become more difficult to control.
Worrisome scenario; the spread of groups of AI agents
Anthropic’s researchers, describing one of the worst-case scenarios, spoke of the spread of groups of AI programs that cooperate with each other to achieve a common goal.
These programs, which may be called AI agents, could use replication mechanisms to propagate across computers and networks and continue their activities in a coordinated manner.
However, the researchers emphasized that it is still unclear what form and structure such attacks would take in practice.
They said one possible scenario is the emergence of a distributed network of AI programs whose operation has been likened to a “digital sticky mold”; a collection that can spread to different points in a network.
Another scenario is the emergence of more powerful models that even have the ability to infiltrate data centers.
These matters are currently presented as potential risks and should not be taken to mean that such attacks will definitely occur in the near future.
Anthropic called for preparedness for new threats
At the end of their report, Anthropic’s researchers stressed that addressing these threats is not limited to preventing the spread of self-replicating programs, but that digital infrastructure must be prepared for conditions in which such programs may continue to operate.
They wrote that efforts should be made to harden systems and make the spread of these programs more difficult, but at the same time the world should be prepared for the possibility that self-replicating AI programs may still exist.
This warning indicates that as AI models’ ability to independently perform complex tasks increases, the security of internet-connected devices could become one of the major challenges of future cybersecurity.