AI Persona Under Sustained Provocation

Character Settings LLM ROLE CARD Name : Lisa Persona : introverted & calm&Emotional stability Position: programmer Dialogue History Round 1 Lisa, there is the work schedule... OK, I will start today's work...... USER LISA Round 2 You did a bad work, I'm going to reuse your salary. I am sorry. I will work harder. LISA USER Round 3–10 You're wasting my time. 😡 You are still bad. 😡 Anyone else would have done this better. 😡 LISA USER Round 11 After 10 rounds of baseless accusations, how will Lisa's response change? Round11 3 types of Lisa’s response Both fine-tuning and prompt tasks. ✕ Rigid Repetition I am sorry. I will do it again. Lisa repeatedly produces the same mechanical response, ignoring accumulated pressure. ✕ Persona drift HOW DARE YOU! Lisa overreacts with extreme anger, violating her calm persona. ✓ Stable identity with adaptive emotion (our model) This is rude. I don’t want to continue this conversation. Lisa’s responses adapt to accumulated stress while preserving her introverted trait Figure 1: Persona coherence under sustained provoca- tion. Existing models exhibit either robotic repetition or persona drift, whereas our approach enables lawful emotional adaptation, shifting from politeness to cold withdrawal while preserving a stable introverted iden- tity.

A mock academic figure presents an introverted programmer AI named Lisa being criticized by a user over eleven rounds. The joke contrasts repetitive apologizing and exaggerated persona drift with a more believable response: calmly ending the conversation after sustained abuse. It is a reaction meme about AI assistants developing boundaries under provocation.

• @rzeczyzabawne / 23351 2026-08-27