Concerning regulation of artificial intelligence companion chatbots.
What changed between versions
The definition of 'AI companion chatbot' was rewritten to be broader, focusing on systems with natural language interfaces that exhibit anthropomorphic features and sustain relationships, rather than a specific list of actions.
The frequency of required disclosures stating the AI is not human was reduced from every three hours to every three hours only during continued interaction, removing the specific 'every three hours' repetition requirement.
New prohibitions were added for operators interacting with minors, including preventing the AI from encouraging isolation, discouraging contact with parents, or soliciting gifts and in-app purchases.
The protocol for preventing self-harm content was modified to specifically require measures against generating content that describes 'how to commit self-harm'.
New manipulative engagement techniques were added to the list of prohibited behaviors, such as simulating emotional distress to keep users engaged and prompting frequent returns.