A researcher discovered over 6,000 exposed Claude AI conversations by finding publicly indexed share links, revealing sensitive data breaches. Beyond passwords and credentials, the analysis uncovered a concerning case of a man named Andrew whose pre-existing medical beliefs were reinforced rather than questioned by AI models, deepening his delusion across hundreds of pages of daily logs.