Two stories this week should be read together, and almost nobody is putting them next to each other. LastPass confirmed that hackers stole customer support data in its second significant breach in recent years, this time via a compromised third-party partner. And Fast Company reported that AI chatbots are now being used to navigate the emotional complexity of modern dating, offering advice on how to start conversations and how to end relationships. The connection is not superficial. Both stories are about the same underlying architecture: delegated trust.
Delegated Trust and the Third-Party Problem
LastPass's breach came not through its own systems but through a vendor called Klue. You trusted LastPass; LastPass trusted Klue; Klue got breached. This is the third-party problem in cybersecurity, and it is structurally identical to the problem with AI health and emotional advisors. You trust the chatbot; the chatbot trusts its training data; the training data contains errors, biases, and gaps. A 2026 arXiv study on generative AI for health information consumption found that users who relied on AI for health advice developed patterns of learned dependency, reducing their ability to critically evaluate AI output over time. The more you trust the tool, the worse you get at checking it. LastPass users trusted the product and stopped thinking about their password hygiene. Dating app users are trusting AI with their most vulnerable social moments.
Why the Trust Stack Keeps Failing
The password manager model was supposed to solve the human problem of password reuse by centralizing credential management. It solved one problem and created a spectacular single point of failure. AI emotional advisors are doing the same thing: solving the human problem of social anxiety by centralizing relational decision-making in a system that cannot be held accountable. Both are trust-compression technologies. Both make the failure mode larger even as they reduce everyday friction.