IT & Microsoft cloud for legal

The reason it keeps working is that it exploits process rather than technology. So the countermeasures have to be both. On the technical side: strong email authentication so the firm's domain is harder to spoof, anti-impersonation controls that catch lookalike domains and display-name spoofing, and multi-factor authentication so a compromised mailbox is much harder to obtain in the first place, since many of these attacks begin with a genuine mailbox rather than a forged one.

On the process side: bank details communicated through a channel that is not email, a rule that details are never changed on the strength of an email alone, and verification by a phone call to a number held on file rather than one supplied in the message. Firms that have avoided losses generally credit the process rule rather than the technology, and clients should be told about the rule at the start of the matter, before an attacker gets the chance to set the expectation instead.

In practice that translates into a recognisable set of things. Confidentiality of client information, which means controlled access, encryption and a defensible position on who can see what. Integrity and availability of matter data, which means backups that have been tested rather than assumed. Competence and supervision, which extends to understanding the risks in the technology the firm relies on. And prompt reporting when something goes wrong, which is difficult if nothing is monitored closely enough to notice.

Cyber Essentials is a useful backbone here precisely because it is externally defined. It does not discharge a regulatory obligation on its own, but it gives a firm a structured, assessable baseline across the five control areas, and evidence that an independent party looked. For firms that also need to satisfy client questionnaires and insurers, one exercise tends to serve all three audiences.

A workable model separates active matters from closed ones. Active matters live where people work, with permissions scoped to the team on the matter. Closed matters move to an archive with retention applied according to matter type, so material is disposed of when the obligation ends rather than when someone remembers. Retention labels in Microsoft 365 can enforce that automatically, which matters because manual retention is retention that does not happen.

Two practical benefits follow. Subject access requests and disclosure exercises become tractable, because the search space is defined rather than the whole history of the firm. And the risk profile drops, since data you no longer hold cannot be breached.

The questions worth answering in advance are specific. Where does the data physically live, and is it backed up independently of the vendor's own infrastructure? What is the contractual recovery time, and has anyone tested whether it is achievable? Can the firm extract its own data in a usable format without the vendor's cooperation? If the system is cloud-hosted, does the vendor's backup protect against the firm's own mistakes, such as bulk deletion, or only against their infrastructure failing?

The recurring gap is assuming a cloud vendor's resilience covers scenarios it does not. Vendor infrastructure redundancy is not the same as a restorable backup of your data, and Microsoft 365 works the same way, which is why an independent backup of email and documents is worth having alongside whatever the practice management vendor provides.

The first distinction that matters is where the data goes. A consumer AI service processes what is submitted on terms the firm has not negotiated and cannot evidence to a client. An enterprise service operating within your own Microsoft 365 tenant keeps material inside the tenant boundary and inherits the permissions already in place, which is a fundamentally different confidentiality position and a far easier one to explain to a client or the SRA.

The second is permissions, and it is where firms get caught out. An assistant that answers from what a user can already access will surface anything over-shared, so a firm with a legacy file share migrated with broad permissions should expect it to find material a fee earner should not see. That is a pre-existing problem being revealed rather than created, but it will be revealed on the day the tool is switched on.

The practical route is an acceptable-use position stated clearly enough that people know what is and is not allowed, an approved tool that operates inside the tenant, permissions and sensitivity labelling reviewed before rollout, and a rule that AI output is checked by the responsible fee earner. That last point is professional obligation rather than technology: the duty to the client does not move because a draft was machine-generated.