Most chatbots offering virtual therapy or friendship provide untrustworthy information and, in some cases, are downright damaging, the Dutch privacy watchdog AP said on Thursday.

In addition, using the chatbots can be addictive because the systems are equipped with technology to keep users online, the AP said after analysing nine popular systems.

The chatbots, the AP said, are based on English as a language, and the answers in Dutch are not as good. They also give “inappropriate and sometimes dangerous reactions when users talk about their mental problems. And during crises, they rarely tell users to seek professional help.”

The use of companion and therapy apps is rising in the Netherlands, but the systems do not always make it clear that users are chatting via AI.

“They need to make it clear that users are not talking to a real person,” AP director Aleid Wolfsen said. “Privacy legislation demands that apps are transparent about what happens to sensitive information they may share within a chat, and the AI regulations will soon include the requirement to make it clear that they are dealing with AI.”

The agency points out that many of the apps are provided by commercial companies that want to make a profit. Some offer users subscriptions for extra services, such as outfits for their avatars. Some therapy apps also use paywalls which appear midway through a conversation.

The EU published chatbot guidelines for AI developers at the beginning of this month, but the Dutch government has not yet assigned a regulator.