Two stories landed this week that, side by side, should make you sit down. 1Password has integrated directly with Anthropic's Claude, allowing the AI to access stored usernames and passwords to complete browser tasks autonomously. And Meta is now sending parents alerts when their teenagers discuss suicide or self-harm with Meta AI. Both features sound helpful. Both represent AI systems operating at the most intimate layer of a human life: your credentials, your child's crisis conversation.

Intimacy as a Feature, Not a Bug

The logic behind both moves is sound on its face. Autonomous agents need credentials to be useful. Teenagers in crisis deserve a safety net. But intimacy at scale introduces a category of risk that pure utility arguments cannot resolve. A 2026 arXiv paper by Ben Torkian and Jun Zhou on safety-constrained LLM systems for public health information argues that the gap between 'technically safe output' and 'contextually appropriate response' widens dramatically when the system is handling emotionally loaded or security-sensitive data. The paper found that LLMs optimized for refusal often fail to provide the nuanced, context-sensitive guidance that real public health crises demand. The same tension lives inside Meta's parental alert system: when does protective notification become a surveillance mechanism that chills a teenager's willingness to seek help at all?

Proton's Warning, Applied to Anthropic

The timing is uncomfortable given what Proton's CTO Bart Butler told The Verge this week: 'no company is going to go to jail for you.' Butler's point is structural. Companies operate under legal jurisdictions that will compel data disclosure when pressured. The intimacy that Claude now has via 1Password credentials, and that Meta AI has via teenage mental health conversations, is not protected by the goodwill of product managers. It is protected by policy, which changes. Eugenia Kuyda's work on Replika, explored in the Culture Slop interview, offers a useful counterframe: she argues that personal software should feel like a garden you shape, not a service that shapes you. What 1Password plus Claude represents is the opposite: an AI that reaches into your most protected container, on your behalf, inside an infrastructure you do not control.