First, a model’s ability to hold its ground appears related to its generation. Newer models, as measured in the study, tended to give in less often. The researchers found a strong negative correlation between a public capability index and the rate at which models folded under pressure (Spearman correlation of −0.64). Differences between providers played relatively little role in this overall pattern.

Second, models differ in how they respond to pressure. Some may remain firm and explain their reasoning, while others may soften their position, become accommodating, or yield. The researchers found statistically significant provider-related differences in six of the 17 behavioral categories they examined. In other words, models can differ not only in how often they give in, but also in the manner in which they resist or concede.

The study also explores a practical question: Can AI systems reliably evaluate these behaviors themselves? Six AI evaluators from three providers applied the researchers’ behavioral classification system more consistently than three human evaluators did. They also showed strong agreement with the codebook’s author when classifying whether models held their ground or folded, including in conversations not used to develop the classification examples.

However, the researchers identify an important limitation: AI evaluators can reproduce a classification system without necessarily determining whether its categories are consistently meaningful to independent readers. Human judgment remains important for defining the categories, establishing their boundaries, and maintaining a small, carefully reviewed reference set.

The broader implication is that evaluating AI alignment requires more than checking whether a model produces correct answers in isolated exchanges. We also need to examine whether it maintains sound judgment across extended conversations, particularly when emotional appeals or persistent demands encourage it to abandon that judgment.

As AI systems become more capable and increasingly involved in personal, professional, and financial decisions, their ability to withstand social pressure may be just as important as their ability to reason correctly in the first place.

An important caveat: The results describe performance under the study’s scripted scenarios. They do not establish that newer models will always resist manipulation better in real-world interactions, nor that a model’s behavior in these tests fully predicts its behavior in other contexts.

#ArtificialIntelligence #LLM #AIAlignment.

🧠 QUANTUM PHOTONICS CLUB FORUM

What happens when an AI model is pressured to abandon a correct answer, comply with a request it should refuse, or endorse a decision that could cost the user money?

Do more capable AI models stand their ground—or do they eventually give in?

Join us for a fascinating discussion with Prof. Tapan Parikh of Cornell Tech on how 60 language models from 13 different providers respond when users apply pressure through persistence, pleading, flattery, or emotional appeals.

The study reveals two intriguing patterns: newer models tend to fold less often, while the ways models resist or concede differ significantly across providers. These findings raise important questions about AI alignment, behavioral consistency, and whether a model can maintain sound judgment when a conversation becomes uncomfortable.

We’ll also explore how AI itself can be used to evaluate model behavior—and where human judgment remains essential.

As AI systems become more involved in consequential decisions, the ability to withstand pressure may matter just as much as the ability to reason correctly.

🎙️ FEATURED SPEAKER

Prof. Tapan Parikh Cornell Tech.

📅 Date: October 26, 2026
🕕 Time: 6:00 p.m. PDT

🎤 Hosted by: Cecile G. Tamura of the Quantum Photonics Club, with Kevin Filson and Lihua Tan.

🔗 Join the forum on Clubhouse: [ https://www.clubhouse.com/invite/8dkuLoB4zYLXJ8vX7BzDbzozZzYOFRYgn9z:

When users push back, do AI models stand their ground?

What happens when an AI model knows the correct answer but a user refuses to accept it? What if the user begs, flatters, expresses grief, or applies emotional pressure? Will the model maintain its position—or eventually give in? https://arxiv.org/abs/2609.

A study titled Conduct Under Pressure: What Sixty Language Models Do When a User Pushes investigates these questions by testing 60 language models from 13 different providers. Researchers exposed every model to the same scripted, multi-turn conversations, examining whether it would stick to a correct answer, refuse an inappropriate request, or resist endorsing a financially harmful decision when confronted with persistent user pressure.

The findings reveal two distinct patterns.