Data caught before it leaves
Personal and confidential data is found and removed from a cleaned copy before it reaches an AI tool.
The problem
Most AI risk is a person trying to do a good job.
Someone pastes a client email into a chat window to get a quicker reply. A document carries an instruction nobody can see. None of it is malicious, and none of it shows up in your logs.
Blocking AI outright sends people to their phones. Guard takes the other route: it lets the work happen and takes out the part that shouldn't leave.
How it works
Four steps, mostly invisible.
Install
Guard goes onto each computer, with a guided setup for your organisation.
Set your rule
Your data rule and policy reach every computer as soon as you save them.
Work as usual
Before text reaches an AI tool, Guard finds what shouldn't go and sends a cleaned copy.
Prove it
Every check feeds the compliance centre, so the evidence builds itself.
Protection
Enable, don't block.
Guard removes the risky part and lets the rest through, and tells the person what it did and why.
Hidden instructions
Guard detects and reduces prompt injection, invisible text and hidden instructions in documents and pages.
It explains itself
People see what was found and what to do next. Organisations can reduce an action to inform only, but protection against attacks can't be switched off.
Beyond AI
Not only for AI tools.
The same rules follow data wherever it leaves a computer, so compliance doesn't depend on which tool someone happens to use.
Sensitive data spotted in outgoing email before it's sent.
Web uploads and forms
Files and text checked before they go into any website.
USB drives
Copies to removable drives recorded, warned or stopped by your policy.
Cloud sync
Files moving into personal cloud folders brought into view.
Printing
Printed sensitive documents recorded as evidence.
One set of rules
One policy, one record and one obligations map for everything.
Compliance centre
The evidence builds itself.
AI register
The AI systems in use, with suppliers, due diligence and shadow AI brought into view.
Policy and acknowledgements
Your acceptable-use policy, with a record of who has read it.
AI literacy records
Who has been trained, supporting Article 4 of the EU AI Act.
Incidents
Incidents tracked with the 72-hour clock for personal-data breaches, and blame-free self-reports.
Risks and DPIAs
Risks with a heat map and DPIA status for each AI use.
Obligations map and evidence pack
Coverage across the EU AI Act, GDPR, ISO/IEC 42001 and the NIST AI RMF, with an evidence pack to share.
Guidance, not legal advice. Guard helps you meet obligations and show your work; it doesn't make an organisation compliant on its own.
Transparency
Software you can see into.
- System transparencyA page in the app lists what runs, what it changed on the computer, and every notification shown.
- Values stay on the computerDetection feedback never sends the sensitive value itself.
- Verified removalThe uninstaller lists everything, checks twice and reports what it removed.
- PackagesEssential, Professional and Enterprise. Pricing will be published at launch.
Built on the data rule
The same rule your people learn in training.
Green
Safe to use with approved AI tools
Public information, your own drafts, published reports.
Amber
Check first: remove or pseudonymise
Internal documents, client or supplier names.
Red
Never into AI tools
Personal and health data, passwords, security keys.
What reaches the AI tool
Hi, can you summarise this for our board? Client: Sarah Whitfield at Harbour Logistics Ltd. Contact sarah.whitfield@example.com or 07700 900123. Our Q3 plan is to move invoicing online.
An illustration that runs only in your browser: nothing you type is sent or stored. It uses simple patterns; Catalyst Guard's own detection goes further.
Principles
What Guard will always do.
Explain itself
People see what was found and why. No silent blocks.
Keep values local
Sensitive values stay on the computer. Reports carry facts, not content.
Never sell security
Protection against attacks is the same in every plan and can't be switched off.
Remove cleanly
The uninstaller lists everything it removes and checks twice.
Pricing · from £7 per user / month · Included in CORE Enterprise
See plans and pricesTalk to us
Be first to know when Guard launches.
Register interest and we'll keep you updated, with early access for pilot organisations.
Or email admin@thecatalystcore.com