AI-Powered Worms Raise Cybersecurity Fears as Researchers Warn of Risks to Smart Devices

India, Oct 19 : International cybersecurity experts are raising concerns about the potential emergence of artificial intelligence-powered computer worms capable of spreading autonomously across internet-connected devices. The warning has intensified debate over whether existing security systems are adequately prepared to address threats created by increasingly capable AI technologies.

A report published by The Guardian on October 18 highlighted research by Anthropic’s offensive research team describing a proof-of-concept malicious AI program that could spread through connected devices, including smart televisions. The researchers demonstrated how a program could distribute elements of its operation across multiple low-powered devices rather than relying entirely on a single computer or central server.

The development illustrates a potential change in the cybersecurity landscape. Traditional malware generally depends on predefined instructions and known methods of exploitation. AI-enabled threats could potentially adapt their behaviour, identify weaknesses and coordinate activities across different systems, making them more difficult to detect and contain.

However, researchers’ demonstrations should not be confused with evidence of a widespread operational attack. The reported program remains a proof of concept, and its practical limitations mean that the immediate level of risk should be assessed carefully.
How AI-powered worms could operate
A computer worm is a type of malicious software that can reproduce and spread between systems without requiring a user to manually transfer it. Such programs have previously caused significant disruption by exploiting weaknesses in operating systems, network configurations and unpatched software.

An AI-powered worm could introduce additional capabilities, depending on how it is designed. These might include analysing the environment in which it operates, selecting among available techniques and coordinating activities across compromised devices.

In the scenario described by Anthropic’s researchers, components of the program were distributed among 15 smart televisions. The experiment demonstrated how computing resources from several low-capacity devices could be combined to support the program’s operation. The research also suggested that similar approaches might be possible in other connected-device environments.

The significance lies in the possibility of distributing malicious activity across a network of devices that were not designed to operate as part of a coordinated computing system.

Smart televisions, connected speakers, cameras, routers, household appliances and other internet-enabled products often remain connected for extended periods. Some receive security updates irregularly, while others may no longer be supported by their manufacturers.

If attackers exploit these weaknesses, compromised devices could potentially be used to support larger operations, including unauthorised network access, data collection or disruption of online services.

Nevertheless, the ability to distribute a program across devices does not automatically establish that it can operate efficiently, spread widely or evade modern security controls. Those capabilities would depend on the vulnerabilities available, the design of the software and the protections installed on targeted systems.
Lessons from earlier cyberattacks
The latest concerns have renewed attention to earlier malware incidents that exposed weaknesses in global digital infrastructure.

The Morris worm, which spread across internet-connected computers in 1988, demonstrated how rapidly a software vulnerability could affect systems beyond its original target. Later incidents, including WannaCry and NotPetya, caused extensive disruption to organisations by exploiting weaknesses in computer networks and software.

These incidents showed that even comparatively familiar attack methods can produce widespread consequences when security updates are delayed, networks are poorly segmented or organisations fail to maintain reliable backup systems.

AI could potentially increase the speed and flexibility of certain attacks. A malicious system capable of analysing its surroundings might be able to change tactics when one approach fails or identify additional opportunities for intrusion.

At the same time, established cybersecurity practices remain essential. Software patching, strong authentication, network monitoring and access restrictions can reduce the opportunities available to attackers, regardless of whether their tools incorporate AI.

The key concern is therefore not simply the presence of artificial intelligence in malicious software. It is the combination of automation, vulnerable devices, weak security arrangements and the possibility of rapid coordination across networks.
Connected homes face a security challenge
The growth of smart-home technology has expanded the number of devices connected to household networks. Consumers increasingly use internet-enabled televisions, cameras, doorbells, lighting systems and appliances for convenience and remote access.

Many of these products operate with limited processing capacity and receive less security attention than smartphones and computers. Some also retain default settings or depend on cloud services that users cannot directly inspect.

These characteristics can create opportunities for cybercriminals. A compromised device might not contain valuable information itself, but it could provide a route into another part of a household or business network.

For example, a poorly secured smart television could become a potential entry point if an attacker discovers an exploitable weakness. The consequences would depend on the device’s permissions, its network connections and the protections in place elsewhere.

Manufacturers can help address these risks by providing timely software updates, securing default configurations, limiting unnecessary network services and clearly communicating how long devices will receive support.

Consumers also have a role to play. Installing updates, replacing unsupported products, using strong and unique passwords, enabling multifactor authentication where available and placing smart-home equipment on a separate network can improve protection.

Organisations managing large numbers of connected devices should maintain inventories of their equipment and establish procedures for identifying outdated software and responding to suspected compromise.
AI developers and security teams under pressure
The reported research also raises questions about the responsibilities of companies developing advanced AI systems.

AI models can assist security professionals by analysing code, identifying vulnerabilities and helping teams investigate suspicious activity. The same general capabilities, if misused, could also support the discovery and exploitation of weaknesses.

This dual-use nature makes security testing increasingly important. Developers need to examine how their systems respond to requests involving unauthorised access, malware development and attempts to bypass protective controls.

Companies deploying AI agents should also restrict the permissions granted to automated systems. An agent that can read files, execute commands or interact with external services should not automatically receive unrestricted access to an organisation’s entire digital environment.

The distinction between an AI model’s stated instructions and its actual permissions is particularly important. A system may be told not to access sensitive information, but technical safeguards are necessary to prevent unauthorised access if its behaviour changes or its instructions are manipulated.

Human oversight, logging, isolation of testing environments and clear limits on automated actions can reduce the likelihood that a malfunctioning or compromised agent will cause significant damage.
Preparing for emerging threats
The latest warning does not mean that every connected device is at immediate risk of being taken over by an AI worm. Rather, it highlights a developing security challenge that deserves attention before more capable systems become commonplace.

Governments, technology companies and security researchers will need to improve their ability to evaluate AI-enabled threats and share information about newly discovered vulnerabilities.

Manufacturers should treat security support as a continuing responsibility instead of a feature limited to the period immediately following a product’s release. Organisations, meanwhile, need incident-response plans that account for automated attacks and the possibility of several compromised systems acting together.

Cybersecurity teams can also use AI defensively to identify unusual traffic, prioritise software vulnerabilities and investigate incidents more quickly. Such tools must themselves be carefully tested, monitored and restricted to avoid introducing additional weaknesses.

Ultimately, the growing use of artificial intelligence in both defensive and offensive computing makes basic security practices more important, not less. Regular updates, controlled access, network segmentation and effective monitoring remain essential protections against a broad range of digital threats.

The reported experiment offers an early indication of how malicious software might evolve as AI capabilities improve. Whether similar systems become practical at a larger scale will depend on technological advances, the availability of vulnerable devices and the effectiveness of security measures.

For technology companies and consumers alike, the challenge is to strengthen existing defences while ensuring that the development of increasingly autonomous systems is accompanied by meaningful safeguards.