Stop sensitive documents leaving before they're shared.
In plain English: it stops staff pasting things like passports and bank statements into AI tools.
Skjold is a locally operated security overlay for ChatGPT, Claude and other generative AI platforms. It scans documents and messages on the device before anything is submitted, blocks restricted material and retains nothing — so staff can use AI without breaching your data-handling policy.
passport_scan.jpg
1.4 MB · image
Blocked by Skjold
Passport detected — sharing identity documents is restricted by your AI policy. Nothing was uploaded or retained.
Scans locally
Before anything is shared
Blocks restricted data
Passports, bank statements, IDs, proof of address
Nothing retained
No content stored or transmitted
Inspect on the device. Decide before it's sent.
A local checkpoint between your staff and the AI tool — no content leaves until it has passed your policy.
Compose or attach
A staff member types a prompt or attaches a document in ChatGPT, Claude or any web AI tool.
Local inspection
Skjold reads the message and file on the device, before anything is submitted.
Policy check
Content is checked against your organisation's AI and data-handling policies.
Allow, redact or block
Permitted content passes through; restricted material is redacted or blocked with a clear reason.
Nothing retained
No document content is stored or transmitted by Skjold. The scan happens and is gone.
Restricted material, caught before it's pasted or uploaded.
Skjold ships with common sensitive-data categories and can be configured to anything defined in your AI policy.
Passports & ID documents
Passports, driving licences and national identity documents.
Bank statements & financial data
Statements, card numbers, account details and financial records.
Proof of address
Utility bills, tenancy documents and address confirmations.
Client & case records
Confidential client files, case notes and internal records.
Restricted business material
Pricing models, contracts and commercially sensitive documents.
Anything in your policy
Fully configurable — define your own categories, patterns and rules.
Detection supports alignment with your policies; it does not replace your own controls or accountability. Coverage depends on document types, configuration and detection thresholds.
Across the AI tools your teams already use.
Skjold sits in front of web-based generative AI, so protection follows the interaction rather than a single vendor.
Browser extension
Installs into managed browsers and works across web-based AI tools your teams already use — no change to their workflow.
Desktop app / agent
A lightweight local agent for desktop AI clients and file uploads, deployable through your existing endpoint management.
The scan happens on the device — and then it's gone.
Skjold is built to reduce data exposure, not add to it.
Runs locally
Inspection happens on the device. Document content is never sent to Skjold servers.
Nothing retained
No copies, no logs of content. Only a minimal policy-decision reference can be kept if you choose.
You stay in control
Policies, categories and actions are set and versioned by your organisation.
One policy, enforced wherever staff use AI.
Skjold applies the same organisation-defined policy that governs your other AI interactions. When you need an audit trail, permitted policy decisions can be recorded as reviewable evidence — without keeping the sensitive content itself.
- Enforces your existing AI and data-handling policy
- Blocks or redacts restricted material before submission
- Keeps no document content — private by design
- Optional reviewable evidence for compliance teams