Door-in-the-Face Requests and Refusal Behaviour in Large Language Models

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.…
Opening of the original on arXiv cs.AI
Summary
This arXiv paper explores how Large Language Models (LLMs) respond to the "door-in-the-face" persuasion technique, where a large request is followed by a smaller, more reasonable one. Researchers tested Anthropic's Claude models to see if they exhibited similar refusal behaviors to humans when faced with these sequential requests. The study aims to understand the underlying mechanisms of LLM compliance and refusal, offering insights into their decision-making processes.
Why it matters
Why it matters: This research probes LLM susceptibility to social influence tactics, a key area for understanding AI behavior beyond simple instruction following. If models like Claude can be swayed by these techniques, it impacts how they are deployed in user interactions and persuasive applications. Competitors like Google Gemini and OpenAI's models will likely face similar scrutiny. Future work should investigate the robustness of these effects across different model architectures and training data, and explore potential mitigation strategies for unintended compliance.
Related: OpenAI: OpenAI’s next big AI model has ‘entered the AGI era’ · Google: Google’s latest AI weather model gives you no excuse to forget your umbrella · Microsoft: What's new in AI? · Meta: Trump Administration Sides With OpenAI in New York Times Copyright Lawsuit · xAI: Sam Altman "AGI by December"
Rated low: routine. Worth knowing, not worth rearranging your day for.
- CamoDocs: A Poisoning Attack Against Retrieval-Augmented Language Models Using Camouflaged DocumentsarXiv cs.AI · Web · Aug 28, 2026
- Sam Altman :‘AGI in 2026’, just as Models Start to [Mis]Train ThemselvesAI Explained · Web · Aug 27, 2026
- Four major AI models suffer rare overlapping downtimeArs Technica AI · Web · Sep 3, 2026
- Four major AI models suffer rare overlapping downtimeArs Technica AI · Web · Sep 3, 2026
- The mystery is solved... and the answer is 40x cheaper than ClaudeFireship · Web · Sep 1, 2026
- Black Box: The Chatbots | Happy Accident | Ep 3 – podcastThe Guardian AI · Web · Sep 3, 2026
- Claude designs proteins that bind in the labClaude on YouTube
- Claude turns calculus into combustionClaude on YouTube
- Claude codes a sentence traveling through a brainClaude on YouTube
- Automated researchers can reliably mitigate alignment failuresAnthropic Research
- CamoDocs: A Poisoning Attack Against Retrieval-Augmented Language Models Using Camouflaged DocumentsarXiv cs.AI
- How scientists are using Claude to accelerate research and discoveryAnthropic News
Questions people ask
- What is the "door-in-the-face" technique?
- It is a persuasion method where a large, likely to be rejected request is made first, followed by a smaller, more reasonable request that the requester hopes will be accepted.
- What did the researchers investigate?
- They investigated how Large Language Models, specifically Anthropic's Claude, respond to and refuse requests using the "door-in-the-face" technique.
More from arXiv cs.AI 23 more
Page generated Sep 3, 2026. Summaries are Subvolts' own; the story belongs to arXiv cs.AI.



