Skip to content
Weaponized · Serious · 2 sources ·
Picture from the techpolicy.press article
Image: techpolicy.press

Bots With Better Small Talk: ChatGPT Powers a 1,140-Account Twitter Botnet

Ourtake

The bots passed as human often enough that the detection problem is now a coordination-pattern problem, not a content one.

What happened

Researchers identified 1,140 Twitter accounts forming a botnet that appears to use ChatGPT to generate human-like posts, stolen images and replies promoting suspicious websites. Current LLM content classifiers could not tell these accounts apart from human users, though their coordination patterns made them detectable.

Sources

Report a mistake