Microsoft’s Windows Recall has gone through more reversals than almost any feature in the company’s recent history. Announced, pulled after a security backlash, redesigned, delayed, and now rolled out again in 2026 with new safeguards — and it’s still generating headlines about how it stores your activity and who might be able to get to it.
If you use a Windows PC, or you’re deciding whether to enable a feature like it, it’s worth understanding exactly what Recall does, what changed since its rocky 2024 debut, and what the ongoing concerns actually are.
What Recall Actually Does
Recall’s core function is simple to describe and unusual in scope: it takes a screenshot of your screen every few seconds, runs text and image recognition on each one, and stores the results in a searchable local database. The idea is that you can later search your own activity in plain language — “find that recipe I looked at last week” or “what was that document I had open during the call on Tuesday” — and Recall surfaces the relevant screenshot.
To do this, Recall has to capture everything that appears on your screen. That includes web pages, documents, chat messages, emails, banking dashboards, medical portals, and anything else visible at the moment it snapshots. There’s no built-in distinction between “things worth remembering” and “things that happen to be on screen” — it captures the screen, not your intent.
The 2024 Backlash, Briefly
When Recall was first announced, security researchers quickly found that the database of screenshots was stored largely unprotected — readable by any application or user with access to the machine, with no additional authentication beyond a normal Windows login. Within days, researchers demonstrated tools that could extract the entire Recall database, including screenshots of passwords being typed, banking sessions, private messages, and other sensitive material, all in plain, searchable form.
The reaction was severe enough that Microsoft delayed the feature’s release entirely and rebuilt significant parts of it before trying again.
What Changed by 2026
The version of Recall now rolling out includes several changes that directly respond to the original criticisms:
Off by default. Recall is no longer enabled automatically — users have to opt in explicitly, and the setup flow includes an explanation of what the feature stores.
Biometric authentication required. Accessing the Recall database now requires Windows Hello (fingerprint, face recognition, or PIN tied to the device’s secure hardware), rather than being readable by anything running on the machine.
Local storage with encryption. The screenshot database is encrypted at rest on the device, rather than stored as a set of plain, readable files.
Sensitive content filtering. Recall attempts to detect and skip capturing certain categories of sensitive information, such as passwords and payment card numbers, though this filtering is necessarily imperfect — it relies on pattern recognition that can miss content it wasn’t trained to recognize.
These are meaningful improvements over the original design. But they address how the data is protected, not the more fundamental question of what data exists in the first place.
Why the Concerns Haven’t Gone Away
Even with biometric gating and local encryption, security researchers in 2026 have continued to flag issues — and the underlying tension is structural, not just a matter of polishing the implementation.
A database of everything is a single point of failure. Once a tool like Recall has been running for weeks or months, its database is, by definition, a comprehensive record of nearly everything you’ve looked at on your computer — every account dashboard, every private message, every document. Encrypting that database and gating it behind biometrics raises the bar for accessing it, but it doesn’t change the fact that the bar, once cleared, gives access to everything at once. Malware that can operate with the same privileges as the logged-in user — which is the threat model for most malware on a personal computer — can potentially reach the same data the user can.
Sensitive-content filtering is reactive, not guaranteed. Filters that try to skip “sensitive” screens work by recognizing patterns associated with passwords, card numbers, or specific app types. New apps, unusual layouts, and content that doesn’t match known patterns can slip through. The feature’s promise — “we won’t capture your sensitive information” — is a best-effort filter, not a structural guarantee, and security researchers have continued to find gaps.
It captures other people’s information too. Recall doesn’t distinguish between your own data and anyone else’s that happens to appear on your screen — a colleague’s message in a shared document, a family member’s details in a form you’re helping them fill out, a friend’s photo in a group chat. Anyone who uses or shares your computer becomes part of your activity record without being asked.
It’s not just Microsoft. Recall gets the most attention because of its rocky launch, but it’s part of a broader pattern. Several operating systems and AI assistants across phones and PCs are adding similar “remember everything you do” features, framed as productivity or search improvements. The questions Recall raised apply to any feature with this design, regardless of which company ships it.
The Deeper Design Question
Step back from the specific implementation details, and Recall represents a particular philosophy about personal memory tools: capture everything automatically, and let search and AI sort out what matters later.
The appeal is obvious — you never have to decide in the moment what’s worth saving, because everything is saved. But it inverts the normal relationship between a person and their own records. Instead of choosing what to keep, you’re choosing what (if anything) to exclude from a record that exists by default.
That inversion has consequences beyond the immediate security questions. A comprehensive automatic record of your screen activity is also a comprehensive record for anyone who later gains access to your device — through theft, a forced search at a border crossing, a subpoena, or a family dispute over a shared computer. The things you’d most want to keep private are not excluded from an “everything” database; they’re in it by definition, mixed in with everything else.
If You’re Deciding Whether to Use Recall
A few practical points if you’re weighing whether to enable Recall (or a similar feature) on your own device:
- Check whether it’s currently on. Search Windows Settings for “Recall” — in the current rollout, it should be off unless you specifically enabled it.
- If you enable it, understand what “pause” does. Recall typically includes a way to temporarily pause capturing — useful before opening anything you specifically don’t want recorded, but only effective if you remember to use it every time, which is the same “remembering in the moment” problem the feature was meant to eliminate.
- Consider who else uses your device. If your computer is shared with family members, a comprehensive activity record captures their activity too, often without their knowledge.
- Think about device loss or seizure scenarios. A locally encrypted database is protected from casual access, but not necessarily from a determined attacker, a forensic tool, or a legal order compelling you to unlock your device.
Recall Isn’t an Isolated Case
Recall draws the most attention because of how publicly its rollout went wrong, but it’s an instance of a broader pattern rather than a one-off. Several operating systems and AI assistants — across both desktop and mobile — have introduced or previewed features that build a continuous record of on-screen activity to power AI search and suggestions. The framing is consistently similar: a feature that makes your device “remember” what you’ve seen so you can find it again later through natural-language search.
The specifics differ — some process everything locally, some rely on cloud components, retention windows vary, and opt-in defaults vary — but the basic shape of the tradeoff is the same wherever it appears: convenience in exchange for a comprehensive, persistent record of on-screen activity that didn’t exist before. Evaluating any individual feature like this means asking the same set of questions regardless of which company built it — where is the data stored, who can access it, what’s excluded, and what happens if the device is lost, shared, or searched.
For anyone managing a work computer or a device used by multiple people in a household, these features are also worth checking at the organization or family level, not just individually — a feature that’s off by default for new users may already be on for an account that’s been active for years, carried forward through updates.
A Different Model: Deliberate, Not Automatic
The contrast with a deliberate personal storage tool is worth naming directly. daftei doesn’t take screenshots of your activity, doesn’t run in the background capturing what’s on your screen, and doesn’t build a comprehensive record of everything you do. It stores what you choose to upload — photos, documents, voice notes, files — encrypted in transit with TLS 1.3 and at rest with AES-256, accessible from iOS, Android, and the web.
That’s a fundamentally different relationship to your own data: a record that reflects what you decided was worth keeping, not everything that happened to be on screen. It also means there’s no “everything” database to worry about if your device is lost, searched, or shared — because the comprehensive record was never created in the first place.
The question Recall puts in front of every PC user — “do I want a tool that remembers everything by default?” — is worth answering deliberately, not by accepting whatever a feature defaults to. For the things you actually want to keep, a tool built around your choices, rather than automatic capture, avoids the tradeoff entirely.