Topic
Transparency, responsibility, privacy, and the limits of defensible systems.
7 articlesResearch review
A USENIX study shows that understandable deletion options for sensitive AI conversations can achieve more than technically sophisticated privacy promises. This makes data control part of conversation quality.
First publishedAnalysis
In an experiment, the consequences of emotional self-disclosure were comparable regardless of whether participants believed they were interacting with a chatbot or a human. Transparency and impact are therefore not contradictory.
First publishedAnalysis
Two experiments show that chatbot disclosure in critical services can reduce trust and loyalty. When the bot fails, transparency can, by contrast, have a positive effect. Context changes the reaction.
First publishedAnalysis
An overview with more than 370 references categorizes risks as inherent problems, attacks, and unintentional errors. Four V&V pillars connect evaluation, verification, runtime monitoring, and regulation.
First publishedAnalysis
AI companions can convey closeness, encourage personal openness, and at the same time create new dependencies or vulnerabilities. A European assessment shows why transparency, data protection, and product design belong together.
First publishedAnalysis
Expertise, responsiveness, design, brand, perceived risk, and data protection all interact. For sensitive conversations, trust is therefore a property of the entire system.
First publishedAnalysis
A classifier can flag risky phrasing. Whether a conversational system acts safely as a result also depends on data fit, error types, response, monitoring, and clear accountability.
First published