Pay Now
Book a security review

NIST CSF 2.0 · PROTECT

Data Security & Loss Prevention

Most data does not leave in a dramatic theft. It leaves in an email sent to the wrong person with a similar name, a sharing link set to anyone with the link and then forwarded, or a spreadsheet copied to a personal drive by someone about to change jobs. Data loss prevention is about the accidents as much as the deliberate acts, and the accidents are far more common.

NIST CSF 2.0 · PROTECTMicrosoft PurviewSensitivity labelsGroundwork for AI

The problem

Nobody knows where the sensitive material actually is

Client records live in the practice management system, and also in a folder from a migration, and also in the mailbox of the person who ran that migration, and also in a spreadsheet somebody built to make a report easier.

That spread is normal, and it is why blanket rules fail. A rule that blocks anything containing a bank account number stops your finance team working by Tuesday. A rule that blocks nothing achieves nothing. The useful version knows what a document is and applies a proportionate response.

This work has also become the prerequisite for adopting AI. A search tool or a Copilot deployment surfaces whatever a person already had permission to see, which means every over-shared folder that nobody noticed for six years becomes discoverable through a plain-language question.

How we roll it out

Discover, label, then enforce - in that order

Enforcing rules before you know what you hold is the single most reliable way to make staff hate a security programme.

PHASE 1

Find out what you hold

Scanning across mail, document libraries and drives for the categories that matter - identity documents, health information, payment details, contracts. The output is nearly always surprising.

PHASE 2

Agree a small label set

Three or four labels, not twelve. Public, internal, confidential, and one restricted tier if you genuinely need it. Every additional label reduces the chance any of them get used correctly.

PHASE 3

Label automatically where possible

Content that matches known patterns gets labelled without asking anyone. Manual labelling is reserved for the judgement calls, because a scheme that depends on people classifying every document will not survive.

PHASE 4

Warn before you block

Rules start in a mode that tells the person what they are about to do and lets them proceed with a reason. This finds the legitimate exceptions your policy did not anticipate, and there are always some.

PHASE 5

Enforce the narrow set

Only the rules that survived the warning phase move to blocking. Everything else stays as a warning or an alert, which still gives you visibility without breaking anyone.

What gets controlled

The routes data actually takes

ROUTE 01

Email to the wrong recipient

Warnings on external sending where the content is labelled sensitive, and on recipients whose address closely resembles an internal one - the auto-complete error that causes real breaches.

ROUTE 02

Sharing links

Controls on links that work for anyone who has them, default expiry dates, and reporting on what is currently shared externally. Almost every business finds links they had forgotten.

ROUTE 03

Copying to personal storage

Uploads to consumer cloud storage and removable media, restricted for labelled content on managed devices.

ROUTE 04

Pasting into AI tools

The newest route and the fastest growing. Rules that recognise sensitive content going into a browser-based AI service, which is the single most common shadow AI behaviour.

ROUTE 05

Departing staff

Alerting on unusual download or forwarding volume, which has a strong pattern in the weeks before somebody resigns.

ROUTE 06

Over-permissioned storage

Folders shared with the whole organisation years ago for one legitimate reason. Harmless while nobody looks, and immediately significant once search or AI arrives.

Be honest about the options

How much control is proportionate

  Policy document only Labels and warnings Labels and enforcement
Staff know the expectation Yes, in theory Yes, at the moment it matters Yes
Catches the wrong-recipient email No Warns Blocks
Visibility of what is shared externally No Yes Yes
Risk of disrupting real work None Very low Real, needs tuning
Prepares you for AI adoption No Partly Yes
Sensible when Very little sensitive data Most businesses start here Regulated or high-value data

Questions

What businesses ask about Data Security & Loss Prevention

Short answers to what gets asked most. If yours is not here, put it to us directly and you will get the same answer in writing.

Will this stop people doing their jobs?

It can, which is exactly why the warning phase exists before anything blocks. That phase surfaces the legitimate exceptions - the client who insists on a particular file transfer method, the regulator who wants a specific format - so the final rule set is shaped by how your business actually works rather than by a template.

How does this relate to Copilot readiness?

It is the same underlying work approached from two directions. Copilot readiness fixes permissions and over-sharing so that AI does not surface what it should not. Data loss prevention adds the labels and rules that keep it that way afterwards. Doing the readiness project without this leaves you clean on the day of launch and drifting a month later.

What licences does this need?

Basic labelling and email rules are available in the Microsoft 365 tiers many businesses already hold. Automatic classification, endpoint-level rules and the more advanced controls require a higher entitlement. We will show you what your current licensing covers before recommending an upgrade, because the answer is often more than people realise.

Can this detect data leaving through AI tools?

Yes, on managed devices and browsers, for the major services. It is not complete coverage - somebody using a personal phone is outside it - which is why the technical control works alongside a sanctioned alternative that is genuinely good enough that people stop looking elsewhere.

How it is bought

Included in a managed plan

This runs as part of a monthly managed plan - set up, operated and reported on by us, priced per person per month with Microsoft licences quoted separately.

Book a security review

Find out exactly where you stand, before someone else does

We will measure what you have against NIST CSF 2.0, tell you the three things worth fixing first, and put it in writing. No charge, and no obligation to buy anything afterwards.