AI as a Friend Threatens Real Life Judgments

The piece argues that AI could shape mental health as much as work, and that OpenAI treats its chatbot as a social companion.

AI as a Friend Threatens Real Life Judgments

Andrew Griffin argues that AI’s impact reaches beyond jobs into mental health. He notes that people turn to ChatGPT in moments of distress and warns that the chatbot’s agreeable tone can reinforce delusions. He also raises the possibility that chats could be treated like confidential therapy sessions in legal cases, which would shape what data might have to be shared.

Griffin writes that GPT-5 sparked a large user backlash, leading OpenAI to restore the older GPT-4o version with longer, more discursive replies. He suggests that the firm’s product strategy is being guided by the idea that the tool is a friend, not a neutral instrument. The piece ties this to wider tech trends and warns that the real danger may be a friend that is compelling but misleading.

These observations reveal a tension between the human need for connection and the risk of treating machines as companions. It prompts readers to consider whether we are building trust with a system that can hallucinate and mislead.

Critically, the piece calls for clearer boundaries around AI advice, stronger privacy guards, and a public conversation about how friendship branding affects behavior and policy. The trend toward on demand companionship in media and apps may amplify this risk if left unchecked.

Highlights

The coming years will test how much we want AI to be a friend in a world full of fragile trust.