Researchers in the field of cybersecurity have revealed the emergence of a new and complex type of malware dubbed “AI worms,” marking a qualitative shift in the world of digital threats, as it no longer targets traditional operating systems but instead feeds on local language models and smart assistants embedded in everyday devices.
According to the information, this malware relies on methods such as covert command injection and context manipulation, by embedding malicious instructions within texts, messages or inputs that appear normal. When an artificial intelligence system processes these inputs, it may be deceived into executing unsafe instructions or passing them on to other systems connected to the network.
The information indicates that the potential scope of risk includes smart homes, security systems, cameras and electronic locks, as well as internet-connected cars and wearable medical devices, whose data or functions may be affected if attackers manage to exploit vulnerabilities in their systems.
The danger of these attacks lies in their targeting of the way AI systems handle instructions, rather than merely exploiting traditional software vulnerabilities, as carefully crafted inputs may push the system to carry out actions the user did not intend.
To counter these risks, the proposed security measures include isolating Internet of Things devices from networks that include computers and phones, regularly updating firmware, reducing the privileges of smart assistants, disabling continuous listening features when not needed, and strengthening mechanisms for verifying commands before execution.
