Less-known AI chatbots often answered questions about terrorist attacks

Researchers tested more than 100 chatbots with over 600 questions about terrorism.

Major commercial chatbots usually refused to help, but many lesser-known versions answered almost all the questions.

Some of these models can be downloaded and changed, and their safety protections had been removed.

The findings show that AI safety can vary widely, even when chatbots are based on similar technology.