Schneier on Security

Research shows LLMs like Claude generate highly predictable passwords with strong patterns: most start with uppercase G followed by 7, certain characters appear in every password while others never appear, repeating characters are avoided, and entire passwords repeat frequently. One specific password appeared 36% of the time across 50 attempts. Bruce Schneier notes this is unsurprising given how LLMs work, but flags it as a real security concern as AI agents increasingly create accounts autonomously. The broader challenge of authenticating autonomous agents is highlighted as an unsolved problem.