Apple has spent years building a reputation around privacy. “What happens on your iPhone stays on your iPhone” is a marketing line, a legal commitment, and a genuine technical architecture choice. Differential privacy, on-device processing, and the Secure Enclave are real engineering investments — not just slogans.
That’s what made the iOS 18.1 disclosure so jarring.
In early 2025, developer Jeff Johnson published a technical analysis showing that Apple had silently enabled a feature called Enhanced Visual Search without presenting users with a clear opt-in choice. The feature sends data from your photo library to Apple servers. For an update that shipped under the banner of Apple Intelligence, a privacy-first AI platform, enabling server-side photo processing by default was a notable contradiction.
This post explains what Enhanced Visual Search actually does, what leaves your device, what Apple says about the privacy protections in place, and what you can do about it.
What Enhanced Visual Search Does
Enhanced Visual Search is an extension of the Visual Look Up feature that has existed in iOS since iOS 15. Visual Look Up lets you tap on objects in photos — plants, animals, landmarks, book covers, art — to search for information about them. It’s powered by on-device machine learning that identifies objects locally.
Enhanced Visual Search extends that capability specifically to landmarks and points of interest. When enabled, iOS analyzes your photos for recognizable locations and compares those locations against a global index of landmarks. The goal is to give Photos a broader ability to identify places in your library — useful if you travel frequently and want your photos organized by the locations they depict.
The practical difference is in where the processing happens. Visual Look Up identifies objects using models stored on your device. Enhanced Visual Search, for the landmark matching step, reaches out to Apple’s servers.
What Actually Leaves Your Device
This is the critical technical detail, and Apple has been more transparent about it than critics initially suggested — though the disclosure came after rather than before the feature shipped.
When Enhanced Visual Search is enabled, iOS does not send your photos to Apple. What it sends is a mathematical representation of certain image regions — specifically, areas the on-device model has flagged as potentially containing landmarks.
The technical term for this representation is a “privatized neural embedding” processed using homomorphic encryption. Homomorphic encryption allows computations to be performed on encrypted data without decrypting it — Apple’s servers can match the embedding against their landmark index without ever having access to the underlying image or decrypted data.
Apple also routes these requests through an OHTTP (Oblivious HTTP) relay, which separates your IP address from the query. The relay is operated by a third party, so Apple’s matching servers never see your IP address. The server that sees your IP (the relay) never sees your query content.
If no landmark is recognized in the embedding, nothing is stored. If a landmark is matched, the result is returned to your device for display in Photos. Apple’s privacy policy for the feature states that they do not retain the embeddings and do not associate them with your Apple ID.
Why Developers and Privacy Researchers Were Still Alarmed
The technical protections are real. Homomorphic encryption is a genuine privacy-preserving technique, and OHTTP relay architecture is a meaningful improvement over direct requests. Apple’s approach is more privacy-conscious than what most apps do when they need to match user data against a remote index.
The controversy was about disclosure, not the technical implementation.
Enhanced Visual Search was enabled by default for most users who updated to iOS 18.1 or macOS 15.1. The feature was not highlighted in the update notes in a way that made clear to non-technical users that photo data analysis would begin reaching Apple servers. Many users who strongly valued Apple’s “on-device” positioning — and who had chosen Apple products partly for that reason — discovered the feature only after developers published analyses.
For users who had understood their photos to be entirely local and private, the discovery that anything derived from their library was leaving the device — however well-protected in transit — was a meaningful change to a core assumption.
Apple’s on-device AI marketing has consistently implied that personal data stays on the device. Enhanced Visual Search is technically true to that framing in the sense that the raw photos never leave. But it complicates the framing in a way that Apple did not proactively communicate.
The “On-Device AI” Framing Problem
This incident is worth understanding as a case study in how “on-device AI” language can create accurate-but-misleading impressions.
True on-device processing means the entire computation — from input to output — occurs on your device using locally stored models. Nothing leaves. This is how Apple’s Visual Look Up object recognition works, how Face ID matching works, and how most of Apple’s Neural Engine processing works.
Hybrid processing means some part of the computation — typically the part that requires comparing your data against a large external index — is offloaded to a server, with privacy-preserving techniques applied to minimize what the server learns. Enhanced Visual Search is a hybrid model.
Hybrid is not inherently bad. For tasks like landmark recognition, where the index is enormous and changes constantly, hybrid processing with strong privacy protections may be the only practical way to provide the feature while keeping the service reasonably current.
But “on-device AI” as marketing language does not communicate the distinction. Users who chose Apple for on-device processing may not know they need to separately evaluate which features are truly on-device versus which use privacy-preserving hybrid approaches.
The Advanced Data Protection Connection
The Enhanced Visual Search story connects to a larger Apple privacy question that became acute in early 2025: iCloud Advanced Data Protection.
Advanced Data Protection (ADP) is Apple’s opt-in end-to-end encryption for iCloud. When enabled, it extends true E2EE to iCloud Photos, iCloud Drive, Notes, Device Backups, and several other categories. Without ADP, those categories are encrypted with keys Apple holds — meaning Apple can access the content if compelled to do so by a legal order.
ADP requires users to actively enable it. Most users have not. Apple’s own estimates have never publicly put ADP adoption above a small minority of iCloud users.
In early 2025, the UK government issued an order under the Investigatory Powers Act demanding that Apple create a backdoor into iCloud data for UK users — specifically, a mechanism to access E2EE content without user knowledge. Apple’s response was to withdraw ADP entirely from the UK rather than comply. UK users can no longer enable ADP; existing ADP users were required to disable it or lose iCloud access.
The UK episode illustrated two things. First, that ADP’s protection is real — the UK government wanted it gone precisely because it worked. Second, that server-side encryption, which covers iCloud for the majority of users who haven’t enabled ADP, is subject to government access in ways that E2EE is not.
Enhanced Visual Search operates independently of ADP — the feature involves landmarks in photos, not the photos themselves going to iCloud. But together, these stories paint a picture of Apple’s privacy architecture that is more nuanced than the marketing communicates: strong defaults in some areas, opt-in protections in others, and hybrid processing for features that require remote data.
How to Disable Enhanced Visual Search
If you want to disable the feature after reading how it works:
On iPhone or iPad: Settings → Apps → Photos → Enhanced Visual Search → toggle off
On Mac: System Settings → Apps → Photos → uncheck “Enhanced Visual Search”
The setting is per-device and needs to be disabled separately on each Apple device you own.
Once disabled, iOS will not send any image embeddings to Apple’s servers for landmark matching. Existing data is not retained according to Apple’s policy, so there is nothing to request deletion of — but you can confirm this by reviewing Apple’s privacy report for the feature in their transparency documents.
The Broader Implication for Photo Privacy
Enhanced Visual Search is a narrow feature with specific privacy protections. But the episode it sparked is worth taking seriously.
The assumption that photos stored on your device are entirely local is increasingly difficult to sustain. Modern smartphone cameras and photo libraries are deeply integrated with cloud services, AI features, and remote processing pipelines — often in ways that are not communicated clearly in plain language to users.
On iOS, the full picture of what your photo library is used for now includes: iCloud sync (photos sent to Apple servers), Shared Albums (available to invitees), Memories and Featured Photos (on-device ML), Visual Look Up (on-device ML), Enhanced Visual Search (hybrid — encrypted embeddings to Apple), and any third-party apps with Photos library access.
Each of these involves different privacy trade-offs. None of them are described in the same place or in comparable terms.
What Genuinely Private Photo Storage Looks Like
If you want a photo library that doesn’t contribute to any server-side processing pipeline — for features you may not use and were not clearly told about — the clearest approach is a dedicated storage service that documents exactly what happens to your files.
daftei stores photos and files with TLS 1.3 in transit and AES-256 encryption at rest. Photos are not analyzed for landmarks, not used to improve Apple’s or anyone else’s AI features, and not processed for advertising. The privacy policy is explicit: daftei never trains third-party AI on user content and never sells data.
The limitation is the same one any non-Apple storage service has: you lose the seamless iOS integration that makes iCloud and Photos useful. That trade-off is real and worth acknowledging. But for users who discovered Enhanced Visual Search by accident and want to think more carefully about where their photo data goes, the trade-off may be worth making.
The Right Question to Ask
For any new feature that involves your photos — whether it’s a new Apple AI capability, a third-party camera app, or a cloud storage service — the question to ask is not “does this company care about privacy?” Most say they do.
The question is: where does the computation happen, who can access what at each step, and did I actively agree to this or was it enabled for me by default?
Enhanced Visual Search has reasonable answers to the first two. On the third, Apple had room to do better.