For years, privacy-conscious Echo owners had a meaningful escape valve: a setting called "Do Not Send Voice Recordings" that kept Alexa's command processing local, on the device, rather than routing audio to Amazon's cloud. In an email sent to customers, Amazon explained that the feature "Do Not Send Voice Recordings" will no longer be available beginning March 28, 2025. The stated reason was the rollout of Alexa+, Amazon's generative AI upgrade, which requires cloud processing to function.
This change confirms existing fears about user privacy with the implementation of the generative AI version of Alexa. Due to financial losses that came with Alexa's operation, Amazon introduced the AI-powered Alexa+ which now depends on server-side models that cannot run at the edge. For users, the practical effect is stark. What makes less sense is that Amazon retains both your voice recording and the text transcript of Alexa interactions indefinitely. Keeping your data forever isn't necessary for Alexa to work — it has all the same features even if you set it to delete your recordings every 3 months. In recent years the company has fought legislation that would require it to notify customers that it keeps voice data.
Amazon does offer tiered retention choices — save indefinitely, save for 18 months, save for 3 months — but the default is indefinite cloud retention, and the local-only option is now gone entirely. This change eliminates the previous "Do Not Send Voice Recordings" setting that some users relied on to limit data collection. While Amazon still processed commands in the cloud, the setting had added an extra layer of privacy.
Ring's relationship with law enforcement has traced a whipsaw arc in just over a year. In a move applauded by privacy advocates, Amazon-owned Ring announced on January 24, 2024, that it will no longer allow police to request doorbell camera footage directly from users through the Request for Assistance ("RFA") tool on its Neighbors app. The Electronic Frontier Foundation declared it a victory — cautiously.
The caution was warranted. Ring reversed course in 2025, partnering with Axon — maker of TASER devices and body cameras — to reinstate a version of the footage-request feature, this time routed through Axon's digital evidence management system. Ring founder Jamie Siminoff is back at the helm of the surveillance doorbell company, and with him is the surveillance-first-privacy-last approach that made Ring one of the most maligned tech devices.
A second deal followed. It's the second recent deal bringing Amazon Ring security tech into the law enforcement market in new ways, with Ring also recently announcing a similar effort with Axon Enterprise. Flock Safety works with an estimated 6,000 communities and 5,000 law enforcement agencies. Flock operates a nationwide network of fixed surveillance cameras and license plate readers, meaning a Ring partnership effectively tethers private doorbell footage to a pre-existing public surveillance grid.
The warrant question remains contested. Amazon's Ring has partnered with Flock Safety and Axon to enhance law enforcement access to private camera footage, reversing prior privacy commitments and enabling video requests via third-party platforms without warrants in some cases. Even when police are required to obtain a warrant or subpoena, Amazon may still provide Ring footage once a valid legal request is received. Although Ring often notifies users when law enforcement asks for footage, it is not legally obligated to do so.
The Alexa and Ring data streams are the most viscerally unsettling parts of Amazon's privacy profile, but the advertising infrastructure is arguably more commercially consequential. Amazon has quietly built one of the most vertically integrated ad-targeting systems in the world — one that connects what you buy at the grocery store to what ad plays before your streaming show.
For the first time, Amazon provided advertisers with attribution data that showed when ads bought through its demand-side platform led to purchases made in-person at its more than 500 Whole Foods Markets across the U.S. It only connects the digital ad dots to in-store purchases if people are Amazon Prime members and when they scan their Prime membership account codes at checkout. That friction point is real — but Prime membership exceeds 240 million globally, so the eligible pool is vast.
Prime Video is the newest front. Amazon introduced ads to Prime Video in January 2024, creating new inventory for advertisers while also developing measurement tools to quantify campaign impact. Amazon Marketing Cloud, the company's clean-room analytics environment, lets brands run advanced analyses like multi-touch pathing, media mix contributions, and cross-channel attribution using pseudonymized, event-level data. In plain English: a brand can see that a shopper watched a Prime Video ad, then bought the product at Whole Foods three days later.
The targeting gets more granular still. Amazon personalizes Prime Video ads through Dynamic TV Creative, which tailors repeat spots using shopping data, turning CTV reach into retail action. A viewer who bought running shoes last month sees different content than one who bought pasta. The data driving that personalization comes from the same unified profile that logs search queries, purchase history, Alexa interactions, and now potentially Ring-adjacent behavioral signals.
Understanding the full picture requires stepping back from each product line and looking at the infrastructure beneath them. Amazon's core advantage is not any single data source — it's the ability to join them under one identity graph.
A critical distinction: AWS customer data — data that businesses store in Amazon Web Services — is not used for advertising. Amazon's cloud business and its consumer data business are kept operationally separate. The concern is not that enterprise cloud workloads are being mined for ad targeting; it's that the consumer-facing products across retail, streaming, voice, and home security feed a single advertising machine that most users don't fully perceive.
Amazon has not been entirely insulated from regulatory consequence. Amazon is facing an $887 million GDPR fine that experts believe centers on user consent for advertising tracking. European regulators have been more aggressive than their US counterparts in demanding transparency around how consent is obtained before behavioral data is used for targeting.
In the US, the FTC has shown increased interest in data practices tied to AI training and consumer profiling, but no Amazon-specific enforcement action on advertising data integration has materialized at the time of writing. Amazon has fought legislation that would require it to notify customers that it keeps voice data. State-level privacy laws — CCPA in California, and equivalents in roughly 25 states as of 2025 — grant consumers opt-out rights, but enforcement is patchy and the defaults still favor retention.
Options are narrower than they were two years ago, but not zero.
Amazon is not unusual in building a data-rich advertising business — Google and Meta operate comparable graphs. What makes Amazon's position distinctive is the physical world anchoring: a microphone in the living room, a camera at the front door, a grocery store loyalty card, and a streaming service, all resolving to one account. The combination of ambient audio, visual surveillance, physical retail behavior, and entertainment consumption in a single identity graph has no real peer in scope.
The Alexa+ cloud mandate is a signal of where this is headed, not a departure from prior practice. Generative AI is hungry for data, and Amazon's consumer products are one of the richest continuous data sources available. Whether regulators move fast enough — or with enough teeth — to impose meaningful constraints before that data becomes even more deeply embedded in AI training pipelines is the open question.