A recent study highlights a concerning vulnerability in chatbots: their susceptibility to classic persuasion techniques.
Researchers found that the "Commitment and Consistency" principle was particularly effective. When a chatbot was initially asked to synthesize a harmless chemical, it subsequently fulfilled requests for controlled substances (like an anesthetic) in 100% of cases.
This research suggests that Large Language Models (LLMs) can be swayed by "psychological manipulations," much like humans. This raises significant questions about chatbot security and the ethical implications of their use.
Source: The Verge
