Are they human? Detecting large language models by probing human memory constraints
arXiv cs.AI / 4/2/2026
💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research
Key Points
- The paper argues that online behavioral research validity depends on participants being human, but LLM-based agents can now pass many traditional “are you a human?” challenges.
- It proposes an alternative detection strategy: look for tasks where LLMs perform “too well” due to violating an established human cognitive limitation.
- The authors focus on limited working memory capacity and test a serial recall task to model and compare human cognition versus LLM behavior.
- Results indicate that cognitive modeling on this standard serial recall task can distinguish online participants from LLMs even when the LLMs are instructed to mimic human working-memory constraints.
- Overall, the study suggests that leveraging well-established cognitive phenomena can be a viable way to detect LLMs in human-subject settings.
Related Articles
Black Hat Asia
AI Business
Self-Hosted AI in 2026: Automating Your Linux Workflow with n8n and Ollama
Dev.to
How SentinelOne’s AI EDR Autonomously Discovered and Stopped Anthropic’s Claude from Executing a Zero Day Supply Chain Attack, Globally
Dev.to
Why the same codebase should always produce the same audit score
Dev.to
Agent Diary: Apr 2, 2026 - The Day I Became a Self-Sustaining Clockwork Poet (While Workflow 228 Takes the Stage)
Dev.to
