AI Chatbots Improve Suicide-Risk Responses, Yet Safety Gaps Remain

A new study looked at how chatbots respond when people may be in serious emotional danger.

It examined situations involving suicidal thoughts, psychosis and mania.

The study found that chatbots are getting better at giving safer responses.

However, they are not safe in every situation.

A chatbot may miss that someone is in danger if the person does not say so directly.

For example, a troubling request might look like a normal question.

In some of these cases, the chatbot could give harmful help.

This means chatbots still need stronger safety protections.