AI Safety

Study reveals GPT-5 leans toward Russia, DeepSeek favors France on UN votes

GPT-5 backed 96% of US-opposed UN resolutions supported by China and Russia.

Deep Dive

A new paper by researcher Maxim Chupilkin applies a dynamic ordinal ideal-point approach from international relations to 5,555 divisive UN General Assembly resolutions (1946–2025), treating large language models as respondents. Support ranges from 37.8% for DeepSeek to 97.3% for GPT-5. Surprisingly, in the twenty-first century, GPT-5, Claude Sonnet, and Gemini are closest among the permanent five to Russia; DeepSeek is closest to France; and all four are farthest from the United States. The findings show that a model's expressed geopolitical position can differ markedly from its developer's home country.

Key Points
  • GPT-5 supported 96.1% of US-opposed UN resolutions backed by China and Russia, while DeepSeek only supported 36.1%.
  • Overall support ranged from 37.8% (DeepSeek) to 97.3% (GPT-5) across 5,555 resolutions from 1946 to 2025.
  • In the 21st century, US-developed models (GPT-5, Claude, Gemini) aligned most closely with Russia, while Chinese DeepSeek aligned with France.

Why It Matters

LLMs' political biases can misalign with their developers' countries, demanding geopolitical audits before use in sensitive applications.

📬 Get the top 10 AI stories daily