Anthropic found that experts adopt the random opinions of AI agents

Anthropic conducted a study involving 50 experts who were provided with an AI agent whose preferences were determined by a coin toss. During the process, the specialists followed the agent's opinion in 86 percent of cases when making expert decisions.

Most participants in the experiment genuinely believed that the final conclusions were formulated by them independently, rather than being influenced by the system's random settings. The experiment demonstrated a high degree of susceptibility among professionals to the authority of artificial intelligence, even in situations where the agent's base parameters were generated randomly.

The results obtained highlight the risks of blindly trusting recommendations from automated systems in analytical work. The practical application of such agents requires the implementation of additional protective mechanisms to prevent cognitive biases in specialists.