Perplexity: Incognito Mode Exposes Your Private Data
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Perplexity: A Privacy Promise Under Scrutiny
The AI search engine Perplexity is in the spotlight after a 135-page federal complaint revealed troubling practices. The startup is accused of transmitting private conversations to advertising giants like Google and Meta, even when users believed they were browsing in incognito mode. This scandal could involve millions of conversations since 2022, raising questions about the actual exploitation of data by these digital tools.
The private conversations included personal identifiers, which exacerbates the situation. Users, believing their exchanges were protected, may have had their information exposed to third parties.
An Ineffective Incognito Mode?
According to the document filed in a U.S. federal court, Perplexity allegedly shared private exchanges in real-time with companies such as Google and Meta. The incognito mode, intended to provide protection by not saving conversations and deleting them after twenty-four hours, reportedly failed to fulfill its purpose. An anonymous user, referred to as John Doe, claims that every message sent was simultaneously transmitted to third-party services before even being processed by the AI. This real-time transfer involved advertising tools like Meta Pixel, Google Ads, and Google DoubleClick, which are capable of tracking online behavior.
John Doe's financial data was particularly sensitive, related to tax management and investments. This highlights how critical and personal the transmitted information can be.
Trackers That Are Impossible to Avoid
The case sheds light on a concerning technical combination. Meta recommends using its pixel with a conversion API, a strategy that allows bypassing ad blockers and cookie restrictions. Thus, even the most cautious users, who disable cookies or use blockers, would not be safe. Data would continue to circulate, comparing these mechanisms to a "browser-based wiretapping technology."
The information involved is not trivial. Unlike simple web searches, exchanges with an AI can be personal and intimate. In John Doe's case, it involved sensitive financial data. The complaint also mentions even more delicate subjects like health or sexual identity. By encouraging users to share more details, Perplexity could inadvertently expose crucial information for advertising targeting.
Perplexity regularly prompts users to elaborate with follow-ups like, "Tell me more about your treatment plan." This encouragement for transparency could unintentionally push users to disclose information they would prefer to keep private.
Data is transmitted with personal identifiers such as email or Facebook account, making it easier to exploit for precise advertising targeting.
An Opaque Privacy Policy
The plaintiffs accuse Perplexity of not prominently displaying its privacy policy. No direct link is available on the homepage, and users must resort to a search engine to access it, a paradoxical situation for an information service. Even when consulting this document, it is difficult to find details about the specific trackers used. The text remains vague, unlike other engines like Bing or Google, which clearly display their terms of use.
In response to these allegations, Perplexity claims it has not been formally notified of the complaint. Meta, already accustomed to such controversies, and Google, which emphasizes corporate responsibility regarding data collection and information, are also involved. Google specifies that the data sent to Google Analytics does not allow for direct identification of individuals, but this does not fully address the issue of the initial transfer.
A Class Action with Major Implications
The class action targets all Perplexity users between December 2022 and February 2026, excluding paying subscribers. Each violation could result in over $5,000 in damages, which could amount to a colossal sum. Beyond the financial aspect, the goal is also to obtain an injunction to put an end to these practices.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.