Amazon's New Policy Change Might Make You Throw Out Your Alexa Device

Amazon introduced a next-gen voice assistant called Alexa+ earlier this year. The new avatar of Amazon's in-house virtual assistant is turbocharged by generative AI. Think of it as the same magnitude of upgrade as the shift from Google Assistant to Gemini. Interestingly, as Amazon prepped with early access testing of Alexa+, it quietly made a worrying policy change. All your voice interactions will now be sent to its servers, and there's no way to disable that exchange.

Amazon's Alexa devices haven't maintained a squeaky-clean history with user privacy in the past. After all, a device that is always connected to Wi-Fi and sends and documents your voice communication to the cloud is never good news; especially considering the terrible track record of companies handling user data. In 2021, Amazon assuaged those concerns to a certain extent, with a new feature that enabled on-device processing of user requests.

It was tied to a relatively lesser-known feature that lets users opt out of sending their voice recordings to the cloud. The feature in question is called "Do Not Send Voice Recordings," and unfortunately, it will officially retire on March 28. With the arrival of an AI-boosted Alexa+ assistant, Amazon is no longer offering that opt-out convenience. The change was not publicly announced, but the Echo device users were notified via email about the shift. Needless to say, the news was not well-received by everyone in the user community, and sparked a heated debate in Reddit forums, raising valid doubts about their privacy with Echo devices.

The argument Amazon is making is actually the same that every other AI company has made recently: They need more user data for training. On occasions, these companies have even overstepped the boundaries of fair usage and copyright laws, landing them in legal troubles and forcing them to ink deals with various publications. Amazon is no different. "As we continue to expand Alexa's capabilities with generative AI features that rely on the processing power of Amazon's secure cloud, we have decided to no longer support this feature," said the email sent by Amazon to users.

One must, however, pay attention to not just the data-hungry nature of AI tools, but also the track record of the companies involved. In 2018, Amazon confirmed that its Echo smart speaker eavesdropped on a couple's conversation and sent it to another person, blaming it on a series of errors. A similar incident transpired with a person in Germany, who was sent 1,700 Alexa audio recordings of a stranger.

In 2023, the Federal Trade Commission and the Department of Justice charged Amazon with violation of children's privacy, levying a multi-million dollar fine and asking it to collect the vast trove of data it has collected, including voice recordings. In 2024, a Surfshark analysis found that Amazon's Alexa harvests 28 out of 32 possible data points on users — the most by any player in the smart home segment, and roughly three times the industry average. With Alexa+, the recording and data collection practices are even more brazen.

When Amazon introduced the "Do Not Send Voice Recordings" tool, it clearly warned users about the importance of sharing data with its servers and the onboard AI processing system. Users who opted out of sending their voice interaction history would lose out on perks such as Adaptive Listening, Live Translations, Dynamic Language Switching, and Reading Sidekick. Moreover, sharing voice data was also considered mandatory if users required access to Alexa Guard, calling, messaging, and voice ID.

To make up for some of those feature gaps, the newer (and more powerful) Alexa hardware shifted to on-device audio processing, allowing a local algorithm to turn the voice commands into encrypted text and feed it to Alexa. On-device audio processing wasn't an entirely offline process, as it still sent encrypted data to Amazon servers, but the audio log was deleted as soon as the processing request was fulfilled. With Alexa+ coming into the picture, we are dealing with a whole different breed of AI.

To be specific, generative AI — which is notoriously power-hungry and requires far more computing resources than a simple language understanding model. Alexa+ is based on generative AI tech, dipping into Amazon's own AI models and a bit into Anthropic's AI stack. To take advantage of the advanced Alexa+ experiences, fairly powerful computing hardware is required. Specifically, a silicone with a beefy CPU, GPU, and preferably, an AI accelerator chip, as well. To give you an idea of what we are looking at, drift no further than Apple Intelligence, which is currently limited to Apple's latest iPhones.