DeepMind Study: LLMs Fail to Generate True Randomness Autonomously

1 reports archived1 independent sourcesupdated 10/27/2026, 02:24:12

Synthesis & Latest Updates

1 Sources Cross-Validated

A new Google DeepMind paper reveals that LLMs cannot generate true random choices autonomously. Experiments show models favor specific numbers regardless of temperature or sampling settings, but larger models succeed when provided with external random seeds. This finding has direct engineering implications for agent security and decision-making mechanisms relying on randomness.

LATEST/A new Google DeepMind paper reveals that LLMs cannot generate true random choices autonomously. Experiments show models favor specific numbers regardless of temperature or sampling settings, but larger models succeed when provided with external random seeds. This finding has direct engineering implications for agent security and decision-making mechanisms relying on randomness.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

Rohan Paul (@rohanpaul_ai)T2Oct 27, 02:24·78 pts
DeepMind Study: LLMs Fail to Generate True Randomness Autonomously
- LLMs exhibit severe bias in autonomous random generation, unfixable by adjusting temperature or top-p
- In simulated security scenarios, LLM pseudo-randomness significantly reduces defense success rates
- Engineering advice: Agents should rely on externally generated random numbers passed into the model, not internal generation
View Related Reports & Analysis Original Source