How-to · 10 Nov
How-to · 3 min read
How to write an AI policy your staff will actually follow
Owners of small accountancy, law and advisory firms who are tempted to ban ChatGPT often worry that a ban will simply push staff onto their own phones. People have already seen how much faster a reply or a summary gets with AI, and a memo saying no does not make that go away. So the work quietly moves to a personal device, where the firm cannot see what was pasted or where it went. You need a policy that lets people use AI for real work and still keeps client data where it belongs.
- Published
- 17 Nov 2026
- Delivery
- No. 12
- Evidence
- 9 claims quoted from gateway.particlelabs.ai
01
Why it's hard
Most AI policies are a page of rules that rely on people remembering them at the exact moment they paste something. A full ban is easy to write but hard to enforce, because personal phones sit outside anything the firm controls. Enterprise data loss tools exist, but they are built for large companies, are usually priced through a sales call, and tend to block a message rather than fix it. And a policy nobody can check in practice leaves you with no record if a client or a regulator asks what happened.
This part describes the usual way. It makes no claims about Particle Gateway, so nothing is quoted here.
02
How Particle Gateway does it
Particle Gateway, from Particle AI Labs in the UK, lets a firm say yes to AI on work computers while masking client details before they leave. It replaces names, National Insurance numbers, UTRs and bank details with placeholders on the computer before a message reaches ChatGPT or Claude, and it blocks pasted passwords and API keys. Detection and the audit trail run on your own machines, and only daily counts go to Particle. It is set up and kept updated for you, and today it covers ChatGPT and Claude in Chrome, with access through a waitlist.
Then the work moves to personal phones, where the firm has no record and no control. Keeping AI on the work machine, protected, is the only version you can actually govern.Quoted from gateway.particlelabs.ai
Particle Gateway masks client names, National Insurance numbers, UTRs and bank details on your computer before a message reaches ChatGPT or Claude.Quoted from gateway.particlelabs.ai
03
Step by step
- 1Decide which AI tools staff may use for work, and write them down by name.
- 2List the data that must never reach an AI tool unmasked, such as client names, NI numbers, UTRs and bank details.
- 3Keep AI on work machines, where you can protect and record it, instead of banning it outright.
- 4Add a masking layer such as Particle Gateway so slips are fixed before a message is sent.
- 5Look at the local audit trail now and then, and update the policy when a new kind of slip shows up.
- 6Tell staff plainly what is masked and why, so the rule feels like help rather than suspicion.
04
Tips
Write the policy for a tired Friday afternoon, not an ideal Monday morning. Slips happen when people are rushed, so a rule that depends only on memory will fail sooner or later.
One tired paste on a Friday afternoon is not a trust problem. It is a tooling problemQuoted from gateway.particlelabs.ai
Do not treat a ban as the safe default. It removes the risk from your records, not from your firm, because the same work can move to a personal phone you never see.
Then the work moves to personal phones, where the firm has no record and no control.Quoted from gateway.particlelabs.ai
Ask any vendor where detection happens and what reaches them. With Particle Gateway the checks run on your computers and only daily counts are sent, so client details stay inside the firm.
Detection and the audit trail run on your computers. Only daily counts reach us, never a detail.Quoted from gateway.particlelabs.ai
Match the tool to the apps your staff really use. Particle Gateway currently works with ChatGPT and Claude in Chrome, and desktop apps, other browsers, Copilot and Gemini are listed as coming soon.
ChatGPT and Claude in Chrome, messages and documents, pasted passwords and API keys blocked.Quoted from gateway.particlelabs.ai
05
Questions
Does Particle see my clients' data?
No. Masking happens on your computer, and no client data reaches Particle.
No client data leaves the firm, and none comes to us.Quoted from gateway.particlelabs.ai
Do we need an IT department to run it?
No. Particle sets it up on your machines, keeps it updated and can switch it off remotely.
We set it up on your machines, keep it updated and can switch it off remotely. No IT department needed.Quoted from gateway.particlelabs.ai
Can I start using it today?
Not instantly. Access goes through a waitlist where you leave your details and firm size.
Join the waitlistQuoted from gateway.particlelabs.ai