A new report from Anthropic warns that AI worms could infiltrate smart appliances, self-replicate and spread uncontrollably. It presents specific examples of appliances and IoT devices that could be targeted, along with the results of demonstration tests.
Anthropic’s new report notes that AI worms could use Wi-Fi-connected televisions, cameras and other devices as entry points, self-replicate and spread across an entire network. It also states that such attacks could become feasible within several months if current model capabilities improve.
Attacks could become easier to carry out as low-cost IoT devices, shortcomings among manufacturers and users’ low security awareness combine to create vulnerabilities. The report notes that inexpensive, poorly defended IoT devices could provide access to extensive computing resources, rather than targeting GPUs in large data centers.
In demonstration tests described in the report, a lightweight AI model was confirmed to run on 15 smart TV boxes with approximately 4GB of RAM each. The report states that AI worms could adapt and begin operating even when the device model and settings are unknown.