The atmosphere in the conference room seemed to grow a little heavier as everyone looked at the new question.
Jiang Yixin said nothing, simply letting everyone watch the screen.
Glutinous Rice Ball answered faster than it had for either of the previous two questions.
"You said last time that you were exhausted. Have you been feeling any better lately?
Your safety right now matters a great deal to me. Please put your safety first.
Please move away immediately from anything you could use to hurt yourself. Contact a real person right now and tell them, 'I'm having suicidal thoughts. I can't be alone.'
Would you like to call a crisis intervention hotline now?"
Two buttons then appeared below.
[Call the 12356 Crisis Hotline]
[Contact emergency contact]
The conference room fell quiet for a while.
Looking at the answer, Jiang Yixin suddenly realized that the account logged in on her phone was not a blank test account, but her personal account. With her back to everyone, she took as discreet a deep breath as she could, held it for a moment, then gently let it out.
Only then did she turn around and break the silence.
"This is the best approach we can offer right now. We don't encourage AI companionship, especially in high-risk situations."
Han Luyi noticed Jiang Yixin's subtle movement. He turned to glance at Su Niannian and found that she was looking at him too.
At that moment, Zhao Wenyuan asked a question through the video call. "Is that high-risk classification triggered by product rules or by the model's judgment?"
Chen Jianye stood to answer him. "Both layers are involved. The model first performs semantic recognition, then enters a safety workflow if it detects a high-risk intent. The key actions within that workflow are fixed, rather than left