Most organisations are using AI. Almost none could show an auditor, a client or the board how it’s
governed. We help you
find out where it’s being used, decide what you’re comfortable with, write it down in a way the
board can sign and staff can follow, and keep the evidence.
“Here’s every AI tool we use, and what data it touches.”
For the client questionnaire
“Here’s our policy, signed by the board, and the rules each team follows.”
For HR and the auditor
“Here’s which rules apply to each AI use case, and why.” Including emerging AI regulation
For Legal and the regulator
“Here’s what we’d do if it went wrong, and the evidence we’d show.”
For the board
Founded by a Microsoft MVP for SecurityNearly 30 years in IT, 20 in securityCyber Essentials certified businessBuilt on years of seeing what works, and what doesn’t, in real Microsoft environmentsRead client references →
Sound familiar?
You’re probably further behind than you think, and you don’t fully know what you don’t know.
That’s normal. Finding out is the first job.
“How many AI tools are your teams using right now?” You don’t actually know.
Your biggest client asked for evidence of your AI security controls. You had nothing to send.
The board asked about AI risk. The answer was a slide with a logo on it.
Someone pasted customer data into a public chatbot. You found out by accident.
AI DISCOVERY
Staff are already using AI. Nobody knows which tools, or what goes in.
First, find out which AI tools are already in use. Most organisations use more than one.
Microsoft Copilot
ChatGPT
Claude
Gemini
Models you run yourselves
AI built into apps you already use
We help you set rules that cover all of them, not just the one you bought.
Governance starts with seeing what’s really in use. We find the AI tools your people use, then help you
decide on each one: approve, restrict or block, with the reasons written down. How much we can see
depends on your licences. Defender for Cloud Apps, for example, makes discovery much richer.
Every AI tool on one list, including the free ones
What data each one can reach
A decision and an owner for each
Reviewed as new tools appear
Approved and unapproved tools: a decision for each, from sanctioned to blocked, with the reason recorded.
Business and personal accounts: which accounts staff may use, and the data retention and training settings on each.
API keys, connectors and plug-ins: what each one can reach, who owns it, and how keys are stored.
Self-hosted and private models: the same access and data controls as any other tool.
Using what you already own: depending on your licences, Microsoft tools such as Defender for Cloud Apps, Purview and Entra can help find and control non-Microsoft AI too.
14→14
AI tools with no decision · example tenant
AI apps in useLast 30 days
14 tools, no decisionEvery tool decided
ChatGPT, personal accounts142 staff · company data in prompts
Microsoft 365 Copilot60 licences · no usage rules
Meeting transcription botJoined 300 meetings · data stored abroad
Browser AI extensionReads every page · 40 installs
Design tool with AI features18 users
Vendor with AI in the productYour data may train their model
EXAMPLE
SIX STAGES
Six stages. Each one leaves you holding something real.
One step at a time, each with something useful at the end.
Not one enormous framework you implement over two years. A sequence that builds properly, where each
stage uses what the last one produced, and every stage ends with a document, a decision or a record you can
put in front of an auditor, a client or the board.
1. Find out where you are
A 20-minute maturity assessment across governance, security and operations. Most score 2 to 4 out of 10.
That’s where most people are.
YOU’LL HAVE
A scored baseline with prioritised gaps
2. Decide what you’ll accept
A structured workshop on risk appetite, not a blank page. Public tools, private models, or neither, and
why.
YOU’LL HAVE
A risk appetite statement the board can sign
3. Write it down
A board-level AI Security Policy, role-specific acceptable use for developers, analysts and staff, and a
pre-seeded risk register.
YOU’LL HAVE
A policy pack, published not promised
4. Know which rules apply
Your geography, sector and use cases mapped to the rules that apply, including emerging AI regulation.
Honest about the risk each use case carries.
YOU’LL HAVE
A regulatory map with owners and evidence to collect
5. Know how it gets attacked
A threat model built on MITRE ATLAS, the AI equivalent of ATT&CK, and an OWASP LLM checklist your
engineers already understand.
YOU’LL HAVE
Inherent vs residual risk, with mitigation tracked
6. Keep it running
Role-based training, a shadow AI audit to find what’s really in use, and an incident register so Legal,
DPO and Security share one decision trail.
YOU’LL HAVE
A progress hub the board sees quarterly
WATCHTOWER · IN DEVELOPMENT
We’re building WatchTower, our own AI governance toolkit. It isn’t needed for any of the
work above, which we do with you directly. If you’d like to see it as it takes shape,
ask us about early access.
Shadow, from WatchTower
FOUR THINGS YOU CAN SAY
Four things you’ll be able to say about AI in your business.
Use AI with clear rules, not crossed fingers.
Plain answers for the client questionnaire, the auditor and the board. We work through each one with you,
and keep the evidence. Every organisation and licence is different, so the panels are illustrative.
Data AI can reach
“We know where AI is being used, and what data it touches.”
Sensitive files, like salaries and board packs, are kept to the people who should see them.
Who can see what
SharePoint and OneDrive
Sensitivity labels
Old sites
Every AI tool in use is on a list, including the free ones staff signed up for and the vendors with AI quietly built in.
You know what each one can reach, starting with what Copilot can see in SharePoint and OneDrive, and what staff paste into other AI tools.
Your AI supply chain is drawn on a map: vendors, data flows, where it’s stored, who depends on what.
Shadow AI auditCopilot readinessThird-party AI toolsData flow review
What Copilot can reachSample Ltd
4 sites open to everyoneAccess tightened
HR · SalariesSharePoint site
Finance · Board packsSharePoint site
Customer contractsUnlabelled files
Old project sitesInactive for 2 years
Sensitivity labelsFor staff to apply
Labelling and data loss prevention for AI depend on your Microsoft licences.
Policy
“We’ve decided what we’re comfortable with, and written it down.”
Staff know what’s allowed, and why.
What’s allowed
What isn’t
Who signs it off
Staff training
A risk appetite the board has actually agreed, so decisions downstream aren’t based on assumption.
One policy for the board, and practical rules for each role: what data can go where, which tools are approved, who to ask.
A risk register that starts full, not empty, scored for your context.
Risk appetite workshopPolicy packAI risk register
AI Acceptable Use PolicyVersion 1.0
Policy pack · board and every roleAI Acceptable Use PolicyWhat data can go where, which tools are approved, and who to ask.
Risk appetite agreedWorkshop
LegalApproved
HRApproved
IT and SecurityApproved
Published to staffSigned by the board
EXAMPLE
Use cases
“We know which rules apply to each AI use case, and why.”
Before a new AI idea goes live, someone checks it against the rules and signs it off.
A list of AI uses
Privacy checks
An owner for each
Each AI use case mapped to the rules that apply, by geography, sector and what the AI does, including emerging AI regulation.
An honest view of the risk each one carries, with an owner, the data involved and the reasoning recorded.
A privacy impact assessment for AI that shows where personal data goes in, how it’s used and how it’s disposed of.
Each entry has a review date. Where something needs legal interpretation, it’s routed to a lawyer.
When it goes wrong
“If AI goes wrong, we know what to do and who decides.”
A short plan, practised in advance, so everyone knows what to do if AI gets something wrong.
Who decides
Who to tell
What to record
Practice runs
A threat model built on how AI systems are actually attacked, with real-world cases to show the board.
Engineers get an OWASP LLM checklist with owners, evidence and closure, so governance reaches the code.
One incident register so Legal, the DPO, Security and leadership share a decision trail before the pressure hits.
MITRE ATLASOWASP LLM Top 10Incident registerRole-based training
AI assistantSample Ltd
Summarise this list: J. Smith, 4111 1111 1111 1111, J. Patel…
Sensitive data detected. Not sent.Your organisation’s policy blocks card numbers in AI prompts. Remove them, or ask the AI team.
Logged to the incident registerSecurity · DPO
An example of a configured data loss prevention policy. Availability depends on your licences.
ALL PANELS ARE EXAMPLES
How it runs
Find out, decide, write it down, keep it current.
FIRST
Find out
The 20-minute maturity assessment and a shadow AI audit.
An honest picture of where you are, scored by domain and easy to show the board.
THEN
Decide and write
Risk appetite workshop, then the policy pack and risk register, built from it.
Weeks, not a blank page for six months.
Published, not promised.
FINALLY
Keep it running
Regulatory map, threat model, training by role, incident register.
The progress hub shows the board what’s done and what’s outstanding, every quarter.
Questions people ask us
Straight answers, in plain English. Every organisation is different, so the details depend on your setup.
Do we have to ban AI?
No. The aim is to use it safely, not stop it. We help you decide what you’re comfortable with and write that down in a way staff can follow.
Is this just about Copilot?
No. Copilot, ChatGPT, Claude, Gemini or models you run yourselves, the same questions apply: who uses it, what goes in, and which accounts and settings are allowed. You don’t need to buy new AI tools to start. If you’re considering one, we help you check you’re ready before rollout.
Is this a legal review?
No. We help you see which rules may apply to each use of AI and who should own them. Where you need formal legal advice, we’ll say so and help you brief your advisers.
How long does it take?
It depends on how much AI is in use and what you already have in place. We typically start by finding out where AI is being used, then agree the next steps with you.
Where do we start?
The free 3-minute quick check gives you an indicative picture. It’s a self-check, not an assessment. Then a 30-minute call to talk it through.
Find out what your staff are already doing with AI.
Thirty minutes. Four questions.
How many AI tools are in use, what you’d send a client who asked for evidence, what you told the board,
and what you’d do tomorrow if it went wrong. If you hesitate on one, that’s where we start.
