Security
If you have found a way round this product, or a weakness in the service behind it, we want to hear it from you rather than watch it on TikTok in six weeks. This page says where to send it, what happens next, and what we will not do to you for telling us.
It also sets out, without adjectives, what actually holds the agent in place on a Windows PC — and what does not.
- Version
- 1.0
- Issued
- 16/08/2026
- Published by
- ClearCode Digital Ltd
Covers the Windows agent, the browser extensions, the cloud service, the parent console, the phone app, and this website.
Reporting something
Email support@pcguardian.co.uk. There is no form, no portal and no account to create.
What to include
- What you did, in enough detail that we can do it too.
- What version of Windows, which edition — Home, Pro, Enterprise or Education — and the agent version if you have it.
- What you expected to happen, and what happened instead.
- Whether the account you tested with was a standard account or an administrator.
That last one matters more than anything else in the list. Almost every defence in this product assumes the child is using a standard Windows account, and a bypass that requires administrator rights is a different — and much less interesting — finding than one that does not.
If you are a parent rather than a researcher
Send it anyway, in whatever words you have. “My son got Chrome working again and I do not know how” is a useful report. We would far rather have a vague one early than a precise one after it has spread.
What is in scope, and what is not
In scope
- The Windows agent, its services, the tray application and the local block page.
- The browser extensions for Chrome, Edge and Firefox.
- The cloud service at api.pcguardian.co.uk, the parent console at app.pcguardian.co.uk, the download host at dl.pcguardian.co.uk, and this website at pcguardian.co.uk.
- The phone app.
- Anything that lets a standard Windows account defeat filtering, application control, screen time or the uninstall PIN.
Out of scope
- Attacks that need administrator rights on the PC. The product’s own documentation says an administrator can undo most of it.
- Attacks that need physical access and other boot media. Documented as undefended, in public, before anybody pays us anything.
- Denial of service, traffic flooding, and anything that degrades the service for real families.
- Social engineering of our staff, our suppliers, or a customer.
- Reports produced only by an automated scanner, with no demonstrated impact. Missing headers on a static page are not a vulnerability.
- Anything requiring a machine that is already compromised by other malware.
What we commit to
- We acknowledge. A human reply, not an auto-responder that closes the loop and nothing else.
- We keep you informed. What we found, what we are doing, and when it ships.
- We patch child-level bypasses within fourteen days. That commitment is in our requirements document, not invented for this page.
- We credit you if you want it, and we keep you out of it if you do not.
- We tell customers when it matters. If a weakness meant a family was not protected in a way they had been told they were, we say so. A quiet fix in a release note is not honest reporting.
Testing safely, and what we will not do to you
If you act in good faith, follow the rules below and report to us promptly, we will not bring a legal claim against you for the testing itself, and we will not report you for it.
- Test on a PC and an account that belong to you. Not a real child’s machine.
- Use your own family account, or ask us for a test account and we will make one.
- Do not access, modify or retain anybody else’s data. If you stumble into somebody’s records, stop, and tell us what you saw and how.
- Do not degrade the service for families using it.
- Give us reasonable time to fix it before publishing. Ninety days is the convention and we will usually be much faster; if we disagree about timing, say so and we will talk about it rather than issue a threat.
What actually holds it in place
This is the complete tamper-resistance model for version 1.0.0. Not a summary of it — the whole thing. If a mechanism is not in this table, it does not exist in the product.
| Attack | What stops it | Mechanism |
|---|---|---|
| Stop the service from Task Manager | A standard account is not permitted to stop it, and the peer service restarts it within two seconds. | Service ACL: no SERVICE_STOP, no DELETE for interactive users |
| Kill the watchdog first | There is no first one to kill. Each service watches the other and restarts it. | QueryServiceStatusEx poll · 2 s · shared heartbeat, 3 misses |
| Boot into Safe Mode | Both services are registered to start in Safe Mode, with networking and without. | HKLM\SYSTEM\CurrentControlSet\Control\SafeBoot\{Minimal,Network} |
| Disable it in the Services panel | Requires administrator rights, which the child account does not have. | Standard-account enforcement at enrolment |
| Uninstall from Add or Remove Programs | The uninstaller checks a parent PIN against the service before it will proceed. | argon2id PIN, verified cloud-side · offline fallback is a HOTP code |
| Change the DNS servers | The setting is put back, and traffic to any other resolver is blocked regardless. | Adapter DNS re-asserted every 30 s · WFP block filter |
| Turn on DNS-over-HTTPS | Three mechanisms at once, so defeating one is not enough. | Browser policy off · WFP block on DoH ranges · bootstrap-domain blocklist |
| Unplug the network | The last policy keeps being enforced, with no time limit, from an encrypted local copy. | AES-256-GCM local cache · offline alert to the parent at 30 min |
| Change the system clock | Time is counted against signed server time, not the PC clock. | Drift over 120 s raises a high-severity alert |
| Remove the browser extension | The remove button is not offered; the extension is installed by Windows policy. | ExtensionInstallForcelist · heartbeat every 5 min |
| Boot from a USB stick and edit Windows from outside it | Nothing. We are not running when the machine is booted from something else. | Not defended — mitigate with a firmware password |
What we do not claim
- No special protection from Microsoft. Some security software runs with a deeper level of Windows self-protection that Microsoft grants only to members of its antimalware programme. We are not a member of it, we do not have that protection, and you will not find us implying otherwise on any surface we control. What resists tampering is exactly the table above.
- No kernel driver in this version. Version 1.0.0 ships with nothing in the kernel. Enforcement is user-mode, and it is honest about what that means.
- Not antivirus. We block the addresses of known malware and phishing sites. We do not scan files, and we do not remove anything.
- No unbreakable anything. This product is designed against a curious child with a search engine. That is the actual threat in almost every household. It is not designed against a determined adult with physical access, and no software installed inside Windows could be.
How the software and the service are built
- Passwords and PINs. argon2id, with a 64 MB memory cost, three iterations and four lanes. Neither is recoverable by us.
- Policy is signed. Every policy bundle sent to an agent is Ed25519-signed; the public key is compiled into the agent and an unverified bundle is discarded.
- The offline copy is encrypted. The policy cached on the PC for offline enforcement is encrypted with AES-256-GCM.
- Transport. TLS 1.3 between agent and service, pinned to our issuing certificate authority. An agent’s token is bound to the machine it was issued for, and a mismatch revokes it and raises a critical event.
- The tray application is untrusted input. There is no path through which the local user interface can change policy.
- Logs hold no secrets. No key, token or credential is ever written to the activity log or the audit log.
- Rate limits everywhere. Sign-in attempts, agent heartbeats and access requests are all limited, and the uninstall PIN prompt locks out after five wrong attempts.
Before we release
Three of the release criteria for version 1.0.0 are security criteria.
- An external penetration test of the cloud service and the channel between the agent and the tray application, with no critical or high finding outstanding.
- Every attack in the published bypass table that we say is stopped, verified as blocked on Windows Home, Pro and Enterprise. The one we say is not stopped is not tested for, because it is not defended and we have not claimed it is.
- Antivirus whitelisting confirmed, with Microsoft Defender as a hard requirement.
If those are not met, the product does not ship. That is a shorter sentence to write than it is to live with, which is rather the point of writing it down in advance.