A lawsuit recently filed against Perplexity AI cuts straight into one of the most uncomfortable questions in the current AI boom: what actually happens to the data users share with these systems once the conversation ends (The Straits Times, 2026).
According to the complaint, users interacting with Perplexity’s AI search engine may have had their conversations transmitted to third parties, including Meta and Google, via embedded trackers that activate as soon as someone logs in to the platform. The allegation goes further. Even users who selected “Incognito” mode, expecting a layer of privacy, were not insulated from this data flow.
If proven, that would mark a serious breach of both expectation and policy and expose a deeper structural issue. AI products like Perplexity are increasingly positioned as trusted interfaces. People ask them questions they would hesitate to type into a search bar. They share financial details, personal concerns, business strategies, and half-formed ideas. The interaction feels conversational, contained, almost private. That perception creates a new category of data: high-context, high-sensitivity, and deeply revealing.
The complaint highlights exactly that. One user reportedly shared details about family finances, tax obligations, and investment strategies. This is not browsing data. It is closer to a digital confessional, structured through natural language. If that layer of interaction sits on top of an advertising-driven data ecosystem, the incentives begin to collide.
AI is being positioned as a trusted interface, something closer to an assistant than a platform. It invites openness because it feels responsive, contained, and increasingly intelligent. At the same time, many of these systems exist within an ecosystem shaped by the economics of data, where information flows across networks that support advertising, optimization, and broader platform integration.
Those two models do not sit comfortably together since an assistant implies discretion, but a data-driven ecosystem relies on circulation. The more AI leans into intimacy, the more that contradiction is in stark relief.
The presence of an “Incognito” mode in the complaint brings that contradiction into focus. Features like this do more than alter system behavior. They establish an expectation. When users activate a private mode, they assume the boundaries of the interaction have changed in a meaningful way. If the underlying data pathways remain largely intact, the gap between perception and reality becomes difficult to ignore.
That gap matters because trust in AI systems is cumulative. Each interaction builds a sense of what the system is, how it behaves, and where the limits are. Once users begin to question those limits, the nature of engagement changes. Conversations become more guarded. Inputs become less complete. The system loses access to the very context that makes it valuable.
The Perplexity lawsuit points towards a fork in how AI products evolve from here. One path continues along the trajectory of the existing internet economy, where conversational interfaces feed into broader data ecosystems, even if that connection remains abstract to users. The other path treats AI interactions as fundamentally more sensitive, requiring stricter boundaries, clearer separation, and a rethinking of how data flows behind the interface.
Neither path is purely technical–both are structural. The boundary between interaction and exposure will be one of the most important design choices the industry must make. The boundary between interaction and exposure will be one of the most important design choices the industry must make.
v