Reference
Troubleshooting & FAQ
The problems people actually hit, with the checks that resolve them — ordered by how often each turns out to be the cause.
If you're troubleshooting before you've read anything else, the basics assumed below — what Locke watches, what masking means — are covered in What Locke is. Some gaps are not faults at all but things this release doesn't cover yet; they're listed on Known limitations.
Locke didn't catch something you sent
Work through these in order. Usually the message went somewhere Locke isn't watching, rather than Locke missing it.
- Is Locke protecting this device? On the Dashboard, check Protection status: Desktop Protection, Browser Protection and Detection engine each have their own state. Also check the title bar — if it reads Resume protection, desktop capture is paused.
- Was it a website? Websites are screened by the Locke browser extension. If Browser Protection reads Needs setup, or you used a browser without the extension, see Websites aren't protected below.
- Is the app or site screened? In Controls › Protected Surfaces, find it under Known AI services and check it's switched on. If it's listed under Not screened — unsupported transport or Recognised, not screened yet, Locke can't screen it in this release.
- Was there anything to catch? Send something unambiguous — your own email address is ideal. A message with no recognizable sensitive data correctly produces no detections. Locke screens a message when it's sent, not as you type.
- Is the data type switched on? Check Controls › Sensitive data, and that the word isn't on your Never flag list — see Controls › Sensitive data. Some types are labelled not enforced; see What Locke detects.
- Prove the detector works at all. Open the Sandbox and press Run test on the sample text. If that finds nothing, the problem is the engine, not your configuration — see the next section.
Locke isn't protecting this device
There is no start button: Locke runs on its own. When something is holding it off, start with Support › Why isn't Locke protecting me? and press Check now. It lists what's wrong and the fix for each, in order, and changes nothing itself.
| What you see | Likely cause | What to do |
|---|---|---|
| Desktop Protection isn't running, or Linux shows Finish setting up Locke | Locke's one-time capture setup hasn't been done on this computer — or, on Windows, its certificate wasn't approved during setup. | On Linux, use Finish setting up Locke on the Dashboard or in the self-check. On Windows, choose Review beside Desktop Protection, then Approve certificate in Controls and Yes when Windows asks. |
| Detection engine reads Not answering, or the Sandbox says the privacy engine isn't running | Locke's engine isn't running. Live protection isn't screening either. | Run the self-check. Then check Settings › Protection › Detection models — if the models aren't downloaded, download them. If it persists, reinstall from sonomos.ai/downloads. |
| AI sites and apps won't load, with an alert that AI sites are blocked | Fail-closed is doing its job: Locke's screening can't be reached, so AI traffic is refused rather than sent unscreened. | Run the self-check to see what's down. Only if you must keep working, open a short fail-open window — see Settings › Protection. On a Teams plan only an admin can open one, on their own device; a member's device says the setting is managed by your admin, so tell your admin instead. |
| It worked yesterday and not today | An OS update, or security software quarantining part of Locke. | Check whether your endpoint protection has flagged Locke, then reinstall. |
| None of the above explains it | Something we need to see. | Send it to Support with the self-check result and Copy diagnostics pasted in, and switch on Attach scrubbed logs. |
Websites aren't protected, but desktop apps are
Almost always the browser extension. Web AI sites such as chatgpt.com and claude.ai are screened by the Locke browser extension, not by the desktop app, and a browser without it isn't screened.
- On the Dashboard, check Protection status › Browser Protection.
- If it reads Needs setup, choose Set up — or go to Settings › Protection › Browser Protection — and press Install Browser Protection.
- Each browser needs its own copy. The card lists the other browsers it finds, each with its own Install button.
Teams admins can install the extension on every member's browsers with a browser policy (force-install), so nobody has to remember.
A request or a file was blocked
| What happened | Why | What to do |
|---|---|---|
| A file wasn't sent | Locke reads plain text, code, CSV, JSON, Markdown, log and RTF files, and PDF, Word, Excel and PowerPoint files. It can't read Apple iWork files, archives, SQLite databases, Outlook mail, video or other binary files yet, so it blocks the request rather than guess. Files set to Block are blocked too. | Send the content as text, or check Controls › Attachments for how each format is handled. |
| A large request was refused | A request too large to screen is blocked, not sent unscreened. The message says it's a size limit, not a sensitive-data decision. | Split it into smaller pieces. |
| Cursor or Windsurf stopped working with an AI service | Tools that send prompts in a binary format can't be screened yet, so on AI services Locke screens they're blocked. | Use another AI tool for sensitive work. See Known limitations. |
| Every AI site and app is refused | Locke's screening can't be reached, and the default is fail-closed. | See Locke isn't protecting this device. |
Locke masked something it shouldn't have
- Find the item in Reports › Detection & masking history or on the Activity page and note its data type.
- If it's one word or phrase that keeps getting caught — a product name, a colleague's surname that's also a city — add it to the Never flag list in Controls › Sensitive data. It's still detected and counted, but sent through in the clear. See The Never flag list.
- If it's a name, place or organization and the match was wrong, lower Controls › Strictness and Modalities › Detection strictness from Strict to Standard, or from Standard to Relaxed. Strictness only affects names, places and organizations; card numbers, keys and IDs are masked the same way at every setting.
- If you genuinely don't need the type at all — an Organization Name in work where company names are the point — switch that type off in Controls › Sensitive data.
The numbers look wrong
| Symptom | Explanation |
|---|---|
| The recorded count seems far too low | By default only You typed is counted. Everything your AI tool attaches on your behalf is protected but not tallied. Turn more sources on in Settings › Protection › What counts as a detection. |
| The count jumped enormously | You switched on an extra source — most likely Injected context, which is usually the largest by a wide margin. Nothing about your protection changed. |
| Items masked before sending and Sensitive detections recorded differ | They answer different questions. The first is this session's live capture feed; the second is the stored, all-time history narrowed by your counting settings. Everything recorded was masked, but not everything masked is recorded in that tally. |
| A report disagrees with the Dashboard | Check the report period — a report covers only the period you chose, and Current app session is not the all-time figure. Both honour the same counting settings, so if those changed in between, save the report again. |
| History is shorter than expected | If Automatic deletion is on (Settings › Data & privacy), anything older than its limit has been removed. Reports › Detection history deletions lists every clear. |
A setting is greyed out or refused
You're on a Teams plan and your device is locked to your team's policy. As a member you see a read-only Team policy page instead of Controls, and settings that would weaken protection — pausing, fail-open — say they're managed by your admin. Only your admin can change them; Locke support can't override your organization's policy. See For team members.
Frequently asked
Does Locke slow anything down?
Screening adds a small delay before a message is sent, on the order of milliseconds for ordinary text. A large request takes longer, and Locke can tell you when one will take a while (Screening progress in Settings › Notifications). Images and audio aren't screened, so they add nothing; files Locke can't read are blocked rather than slowed.
Does it work offline?
Detection and masking run on your machine and don't send your content anywhere, so inspecting a message needs no network round trip. The app does need to be online to sign in to your Sonomos account and to periodically confirm your licence (and, on Teams, your seat and policy updates). In practice you're online anyway — the AI tools Locke screens need the internet.
Does closing the window turn protection off?
No. The app is a window onto Locke, and Locke keeps running without it. To actually stop it, use the pause controls — see The window — or, to remove it, Uninstall Locke.
Does signing out turn protection off?
No. Locke protects the device, not the login, and signing out doesn't release your team's policy either: a device your team has locked stays locked on the last policy it received until Sonomos confirms, for that same account, that the lock is lifted.
Will the AI get confused by the masked values?
Rarely. Each value is replaced by a short token that names its type, like EML_x9k2 for an email, and the same value always gets the same token — so an AI can follow references across a conversation. Where it does matter is a task about the value itself: "is this card number valid" can't be answered about a token. For those, use the Sandbox to see what's being replaced, then decide whether to add a Never flag entry or narrow a data type for the duration.
Can I get the original values back?
Your originals never left your machine — you already have them; only the copy sent to the AI was replaced. Nothing turns a token back into the value, and Locke doesn't rewrite what you typed in the AI tool's own interface.
Can Sonomos see what Locke catches?
No. Detection is local, and the values Locke finds are never uploaded. Your history is a local record of metadata, not the values. A support message includes logs only if you switch on Attach scrubbed logs: the newest lines from Locke's app and supervisor logs, with account names, paths, addresses and tokens removed. Detection and screening events are never included.
Can my admin read my prompts?
No — and they don't see your figures either. An admin sees your membership, which team policy your device has applied, and whether you generated a report they asked for — never your reports, your detection counts or anything you typed. The full answer is at the top of For team members.
Can I use Locke on more than one computer?
Yes — install it on each. Detection history and reports are per device, so numbers are per machine too. Your Controls choices and your theme, language and notification preferences follow your Sonomos account to every device you sign in on. On a team, your admin's policy applies to each of your devices.
Is Free really free?
Yes. It's the complete protection engine for one person, not a trial. The one limit is reporting: one Monthly privacy summary a month. Teams adds full reports and team management, not better protection. See Free vs Teams.
Still stuck
Go to Support & feedback, run Why isn't Locke protecting me?, paste in Copy diagnostics, switch on Attach scrubbed logs, and say what you expected to happen. Your version is also under Settings › About & removal. Don't paste the sensitive value itself — describing the kind of value is always enough.
