You can't govern what you can't see.
Tools, plugins and agents arrive before security knows. In our sample, 5 of 6 AI apps were unsanctioned.
Findings from an illustrative assessment of a mid-sized company. Your numbers will differ. The pattern rarely does.
Sample report · illustrative data
| Application / agent | Flow | Primary users | Sanctioned | Severity | Observation |
|---|---|---|---|---|---|
| ChatGPT (consumer) | Web | Marketing, HR | No | High | Customer PII and internal HR policy documents pasted into prompts. |
| Custom pricing agent | API | Engineering | No | Critical | Over-permissive production-database access; raw database structures sent to an external API. |
| Perplexity | Web | Strategy, Sales | No | Medium | Bypasses data-sharing controls; used to summarise unreleased product specs. |
| Claude (API) | API | Development | No | High | Proprietary source code and internal patterns uploaded. |
| Microsoft Copilot | Integrated | Enterprise-wide | Yes | Low | Operating within approved tenant boundaries. |
| Support SaaS plugin | SaaS plugin | Customer Support | No | High | Unvetted plugin reading helpdesk tickets and sending payloads to an unknown endpoint. |
Sample report · illustrative data
5 of 6 AI apps in use were unsanctioned. 42 customer-data payloads reached public models. One internal agent had read access to the full production database.
Shadow AI is shadow IT with a data problem: what leaves isn't a device, it's the contents of the prompt.
Tools, plugins and agents arrive before security knows. In our sample, 5 of 6 AI apps were unsanctioned.
PII, contracts and source code get pasted into consumer AI. DLP was built for files, not prompts.
GDPR fines reach 4% of turnover, EU AI Act fines 7% or €35M.
The EU AI Act starts with knowing which AI systems you run. Few can show who asked what.
Broad database and tool access means one prompt injection can trigger a full table scan.
Subscriptions on corporate cards, API usage unattributed. Finance feels this first.
Up to 4% of global turnover.
Customer PII sent to unapproved models without a DPA.
Source code, pricing logic, unreleased specs.
Competitive advantage leaves through the prompt box.
Duplicate AI subscriptions on corporate cards.
No procurement, no volume pricing, no attribution.
A 30-minute call to agree network segments, departments and data categories. Setup takes a few days.
The gateway runs in passive mode, sees AI traffic, changes nothing and blocks nothing.
Findings, scorecard and remediation plan, presented to your leadership team.
PII leaks rerouted or redacted. Unvetted plugin blocked at the perimeter.
Pricing agent limited to read-only, approved tables. Code uploads scanned and token spend attributed.
Signed audit logging on every request. Full posture verified against the report.
Sample report · illustrative data
See where personal data goes, and stop it reaching non-compliant processors.
A full AI system inventory, ready for risk-tier classification.
A signed record of every request: who asked, what was sent, what came back.
The assessment provides evidence to support compliance programmes; it is not legal advice.
Five yes/no questions. Your score is computed in your browser; nothing is sent until you ask for the result.
/5
Most AI use is likely invisible to you today. A 14-day measurement would show what is running and what is leaving.
You have some controls, with gaps in visibility or enforcement. Measure the gaps before deciding where to spend.
You have the basics in place. An assessment can confirm it against live traffic and cover agents and MCP.
Ready to measure it for real? Book free assessment →
A short form. We reply within two working days to book a 30-minute scoping call.
A network-level gateway in passive mode. No endpoint agents and no application changes. We'll confirm the exact routing (proxy, DNS or egress) with your network team on the scoping call.
No. During the assessment nothing is blocked or altered.
In the EU. Payloads are obfuscated for PII by default in the workbench. A DPA is signed before kick-off.
Only your nominated team and the Smalldata delivery team.
If you deploy AI systems used in the EU, very likely. The report maps your inventory to risk tiers.
You can switch the same gateway to active protection, with critical leak prevention in 24 hours and full posture in 14 days. Or you can stop there: the report is yours.
It's built on 14 days of measured traffic, not interviews and questionnaires, and the enforcement tooling is already in place.
Yes. Reports and briefings are available in Italian or English.