An AI data-loss prevention gateway for regulated industries.

Termon checks every prompt your people — and their AI agents — send to OpenAI, Anthropic, Azure OpenAI or a model you host yourself: secrets blocked, personal data redacted, every decision recorded. Start in monitor mode: within days you see what is actually leaving, without blocking anything.

Book a 15-minute discovery call How a pilot works →

Built in Belfast · Opening a small design-partner programme for regulated and IP-sensitive organisations

Why now

It's already happening — whatever the AI policy says.

88%

of office workers at large organisations have shared work information with public AI tools such as ChatGPT, Claude and Gemini.

PagerDuty Shadow AI Survey, by Wakefield Research: 1,250 office professionals in the US, UK, Australia and Japan. June 2026.

34%

have put customer data or information into them.

The same PagerDuty survey. June 2026.

Nearly half

of enterprise AI use bypasses corporate security.

Akamai, The Enterprise AI Usage Risk Report 2026. August 2026.

The first step is seeing what actually leaves. That's where a Termon pilot starts. How it works →

The risk

"We turned off training" isn't the same as "our data is safe."

Every team rolling out AI hears the same reassurance: the provider doesn't train on your data. It's true — and it misses the point. Turning off training doesn't mean your data stays under your control.

It still lands in their logs

Prompts and responses — PII, source code, contracts — are retained in the provider's logs and backups, reachable by their staff, support tooling, and subprocessors. "Private mode" governs training, not retention.

You can't prove what left

When a regulator or your own board asks which customer data has gone to an AI tool, and who sent it, you have no record. The provider won't give you one.

Breach and legal exposure

Logs get misconfigured, breached, or subpoenaed. Once your data sits in someone else's systems, you've lost control of its lifecycle — retention, copies, and all.

It can be pulled back out

Data held in logs, model memory, or connected tools can be surfaced through prompt injection — not theoretical, demonstrated repeatedly.

The provider protects their systems. Termon protects yours — scanning, redacting, and logging every interaction before it ever leaves your network.

What Termon does

Three problems CISOs and compliance teams keep running into.

Prevent AI data leakage

Real-time scanning of every prompt before it reaches the AI provider. Detection for secrets (AWS, GitHub, Stripe, JWT keys), source code, internal documents, PII, and your own sensitive terms. Block, redact, or warn — per team, per tool, per user. Fails closed: if a prompt can't be checked, it doesn't leave.

Audit trails for compliance

A record of every decision, with the reasoning behind it — metadata only, never prompt content. Built for GDPR accountability today, and for the EU AI Act's record-keeping duties as they arrive. Exportable for regulators. Per-tenant retention. Your compliance team gets the artefact it needs to say yes.

AI agents, not just chat

Coding agents read files and run tools on their own. What an agent reads is checked before the model sees it — a cloud key in a file it opened never reaches the model. What it tries to do is recorded in the audit trail. Tested with Claude Code.

The Termon console's Security posture page: enforcing, failing closed, five detection engines working, what is always removed, and how far protection reaches — including that a streamed answer cannot be redacted. — The console shows, live, what was scanned, redacted and blocked — and the posture page says plainly where protection stops. Demo data.

Design-partner pilot

How a pilot works.

Start by seeing, not blocking. Nothing changes for your staff until you decide it should.

1

Install, together

We stand Termon up with you in your own cloud or on your own servers. Your AI tools change one URL to go through it.
2

Observe — nothing blocked

In monitor mode every prompt is scanned and recorded, and nothing is stopped. No workflow is interrupted.
3

See your own exposure

Within days, a report of what actually left: by category, risk and tool. Metadata only — never prompt content.
4

You decide

Switch on enforcement, adjust the policies, or stop. Your records are yours and exportable either way.

The terms: a fixed window with a clear scope, free or nominal for design partners, under a mutual NDA. You work directly with the engineer who builds Termon.

A sample Termon exposure report for seven days: 153 prompts through the gateway, 66 sensitive findings, 14 blocked, 46 redacted before sending, broken down by category, risk and tool. — The exposure report from step 3. A sample, for illustration — no prompt content is ever stored or shown.

Why Termon

Designed for regulated UK and EU buyers, not Silicon Valley.

Most AI security tools run as SaaS in someone else's cloud. Termon is the alternative for organisations with stricter requirements.

What you get

Drop-in proxy for OpenAI, Anthropic and Azure OpenAI, or a model you host yourself — your tools change one URL
Per-team policies — block, redact, warn or allow — starting in monitor mode
A per-decision audit trail with the reasoning behind each decision, filterable and exportable to CSV
Cost visibility — by team, by tool, by user

How it's different

Runs in your environment — your cloud or your own servers. Termon's own records never leave it and nothing phones home; prompts go only to the AI provider you choose — or nowhere, with a model you host yourself
Fails closed by default — if a prompt can't be checked, it doesn't leave
Already on Microsoft 365? Termon is for organisations that won't route prompts through Microsoft's cloud, aren't on E5, or use other AI providers
A flat organisation-level licence — no per-token or per-request fees

About

Built by an engineer, not a sales team.

Termon is built by Liam Cameron, an engineer with over twenty years building software for government, banking, financial exchanges, and healthcare — regulated sectors where data is handled carefully because it has to be. It's built for that world by someone who has worked in it.

It's at the design-partner stage: working software, 1,450+ automated tests, and real demos — offered to a small number of partners rather than sold broadly. You work directly with the person who builds it.

If you're a CISO, compliance officer, or technology leader at a regulated organisation, and your AI projects are stuck in compliance review — let's talk. Not selling. Researching the right thing to build.

Termon is a trading name of Licam Ltd (UK). Belfast, Northern Ireland.

15 minutes. No pitch.

Tell me how your organisation is approaching AI governance. I'll tell you what I'm seeing across the market. Whether or not Termon ends up being right for you, you'll leave with useful context.

Book a 15-minute discovery call

Prefer to read first? Email info@termon.tech and I'll send the design-partner programme — what a pilot involves, on both sides.