Tech firms redefine recording for AI hardware

Apple states in its Music Detection language that it does not record audio during these processes. However, once a song is detected, one might ask if that counts as a record you heard. The Apple Watch has had a microphone for ages to support Siri and phone calls from the wrist. These new features add complexity to the idea of what data is actually stored on your device.

Smart home cameras may create text snippets instead of saving video footage

Last week, Bloomberg reported that Apple is working on a smart home camera with unique privacy features. This camera does not save traditional video recordings of your home environment. Instead, AI uses the visual data to create text snippets that describe what is happening around you. The idea is that this approach still protects some level of privacy for homeowners. This technology is similar to how cameras in rumored future AirPods will also work.

Facial recognition support is supposed to be available without requiring video footage storage. The company argues this method avoids the legal and ethical issues associated with video surveillance. Users might feel safer knowing no video files are being kept on their devices or in the cloud. This represents a shift from capturing images to generating descriptive text about those images.

Google is exploring image sensors that process visual data without creating records

Google is also mulling this concept of processing visual data without saving footage for posterity. Sandeep Waraich, Google's senior director of wearables, told The Verge about this idea. He suggested cameras on smart glasses could process visual data in your environment using AI. The camera would act as an image sensor rather than a recorder of events. This approach allows the device to see without necessarily creating a record you can review later.

If glasses could process visual data, they might not need to save any footage at all. Waraich argued that in this way, the camera could just be an image sensor. The distinction matters because preserving and reviewing data defines what counts as a recording. Tech companies are testing definitions of recording to make AI hardware more palatable for consumers.

Raw data for these features is deleted immediately after processing in secure areas

For Apple's Audio Intelligence, the raw data is deleted once it finishes processing. This happens in a separate, secure exclave on the S11 processor inside the watch. Any information sent to the cloud goes through private cloud compute systems. These systems are designed so that Apple or third parties cannot access the specific data easily. The company claims this ensures the data is inaccessible after it serves its purpose.

This deletion process happens immediately after the AI completes its analysis task. The system does not keep a copy of the audio or video for future retrieval by the user. This raises questions about whether any transcript generated counts as preserved and reviewable data. If text snippets are created, do they count as records that could be subpoenaed in court?

Why it matters

Privacy laws depend heavily on how we define what constitutes a recording of someone's life. If companies can claim they are not recording you, legal protections for your privacy might change. You might lose the right to review or delete data that was generated by AI systems. The incentives for Big Tech to collect this data are obvious because personal data is money. Money translates into power and influence within society and political systems.

Society must reach a consensus on what constitutes a recording before these technologies spread further. Currently, hardware seems to be the main focus of the surveillance tech debate. Getting rid of cameras solves some problems but does not eliminate all risks. The queasy feeling comes from knowing a preserved snippet of you can exist in someone else's device. This happens even if the user never saw or heard the original event themselves.

Apple and Google are developing hardware that processes data without saving it as a traditional recording

Apple and Google are building devices that handle information differently than standard cameras or microphones. These new gadgets process visual or audio input locally on the device itself. They do not save the raw footage or sound waves onto your phone or cloud storage. Instead, the hardware converts the input into text descriptions immediately after processing. This approach aims to reduce privacy concerns by deleting the original data after use.

The goal is to create a system where no permanent record exists for the user to review later. Apple calls this Audio Intelligence on its smartwatch and upcoming camera projects. Google's team is exploring similar concepts for smart glasses and other wearable tech. Both companies argue that because they do not store the file, they are not recording in the traditional sense.

This shift changes how we think about data collection and storage. It moves away from keeping a log of your life to generating temporary insights. The raw sensory data disappears once the AI task finishes its work. This means you cannot play back the video or hear the audio at a later date.

The technology relies on powerful processors to handle this conversion in real time. Apple uses its custom S11 processor for these secure, local calculations. Google is integrating similar capabilities into its future hardware designs. The processing happens inside a protected area of the chip known as an exclave. This ensures that even Apple staff cannot easily access the specific data streams during processing.

The company claims this architecture ensures the data is inaccessible after it serves its purpose. The deletion process happens immediately after the AI completes its analysis task. The system does not keep a copy of the audio or video for future retrieval by the user. This raises questions about whether any transcript generated counts as preserved and reviewable data. If text snippets are created, do they count as records that could be subpoenaed in court?

How OpenSmartRoute helps

Teams using OpenSmartRoute gain immediate control over data privacy and cost when firms like Apple and Google process audio or video without saving it. The router scores every candidate on quality, cost, speed, and safety while the team sets the weights for each request. Hard rules pin a request so text with personal data stays on an on-premises model and no region or cost cap is ever crossed.

The platform learns from outcomes so a model that answers well gets more traffic and one that fails gets less. A new model is just one catalogue entry and competes on the next request without changing anything else in the app. The hosted platform keeps a models catalogue with prices and public rankings built from real traffic to guide decisions.

An input guard spots prompt injection and personal data before a request leaves the system entirely. 'osr eval' measures routing accuracy on the team's own prompts and can fail a build when it drops. A savings ledger shows what each routed request cost next to what the most expensive model would have cost.

What to do

You should check your device settings to see which features are currently opt-in for you. Look for privacy language in app descriptions that explains how data is handled and deleted. Compare different brands to see who is transparent about their data retention policies. Consider whether you want AI features that generate transcripts or summaries before buying new hardware. Be aware that some devices might have indicators like white lights that signal recording intent.

You can also look at recent news reports from Bloomberg or The Verge for the latest updates. Tech companies are adding cameras to smart glasses and other gadgets in ways consumers have not considered yet. The best October Prime Day tech deals we found often highlight these new privacy features. Keep an eye on how definitions of recording evolve as more AI hardware enters the market. Your ability to control your personal data will depend on these societal and legal definitions.

Source: https://www.theverge.com/column/1005697/we-cant-just-change-the-definition-of-recording

LibreOffice refuses AI by default in new software update

LibreOffice will not include generative AI features in its standard installation.