AI Spend & Risk Report
The AI Spend & Risk Report puts one environment’s AI spend and AI risk for a period into a single document to share. Use it as the deliverable at the end of a 30-day pilot, or as the monthly report of a managed service.
Open it from Home → Reports (/reports). Download it as a PDF (with a contents page and page numbers) or as a single HTML page, or have it emailed to chosen members every month.
Prerequisites
Section titled “Prerequisites”- The Generate the AI spend and risk report permission (
reports.read) to download it. Administrators, Compliance officers, Analysts, Members and Viewers have it. - The Schedule the AI spend and risk report permission (
reports.manage) to change the monthly email. Administrators have it. - For the monthly email, email set up on the deployment (Amazon SES or SMTP, the same transport alert emails use).
What the report contains
Section titled “What the report contains”The cover names the organisation, the environment, the period, the total AI spend, its change on the previous period of the same length, the number of findings and when the report was generated. Then:
| Section | What it shows |
|---|---|
| Executive summary | Total AI spend, its trend, how much is backed by bills, and the top five findings |
| What AI cost | The total and its parts (seats, charges, vendor bills, cloud AI, estimates), the same total Finance shows |
| Spend by tool, model and team | The top ten of each, with share and responses |
| Top spenders | The ten people who cost most, pseudonymised unless the reader may view the collector’s people |
| Vendor reconciliation | What OpenAI and Anthropic reported against Shield360’s estimate, the variance and its reasons. With no vendor billing connected, a note to connect your admin keys |
| Subscriptions and idle seats | Seats assigned, used in 30 days and idle per subscription, and what the idle seats cost a month |
| AI tools discovered | The AI inventory’s counts, and the tools in use that are not approved |
| Shadow AI | Tools used with no tracked subscription |
| Personal AI accounts | How many personal or other organisations’ AI accounts were used on company machines, by tool |
| Data protection and Prompt injection | Counts by kind of data and kind of injection, and how many were blocked |
| Budgets | Each enabled budget’s amount, spend, forecast and status |
| Coverage | Machines, browsers and AI proxies enrolled and reporting, collectors that check in without data, and whether coding agents are checked for sensitive data |
| Recommendations | What to do, most important first, each with the figure behind it |
| The AI analyst’s view | The analyst’s weekly briefing from the period, only when the AI analyst is configured and wrote it |
A section whose source could not be read says so; the rest of the report is still complete.
Recommendations
Section titled “Recommendations”Recommendations follow fixed rules from the report’s own figures, so the same data always gives the same list. For example:
- Idle seats → reclaim them, with what they cost a month.
- A tool in use that is not approved → review it in the AI inventory.
- A rejected tool still in use → block it or record an exception.
- Personal AI accounts on company machines → move those people to company accounts.
- No vendor billing connected → connect your OpenAI and Anthropic admin keys.
- A budget over or forecast over → review it with its owner.
No model writes them.
What the report never contains
Section titled “What the report never contains”The report holds aggregates only. It never contains prompts, responses, files, detected values, account identifiers, email addresses or API keys. Data protection and prompt injection are counted by kind; personal AI accounts are counted. API keys and agents are counted in the inventory, never named.
People are named only in a copy downloaded by someone allowed to view the collector’s people. Everyone else, and every emailed report, sees Person 1, Person 2 and so on, numbered by spend within that report. A budget on one person is shown as Budget on a person.
The report is rebuilt each time it is downloaded or sent, and is never stored. Every download and every email is recorded in the audit log, without the recipients’ addresses.
Choose a period
Section titled “Choose a period”| Period | Covers |
|---|---|
| Last 30 days | The 30 days up to now: the pilot’s deliverable |
| Calendar month | A month in your organisation’s time zone. The month in progress is read up to now |
Email it every month
Section titled “Email it every month”The Monthly email panel shows whether the email is on, who receives it and the month the next email covers.
Tick Email the report every month and choose up to 20 members of the organisation. Auditors and a monitoring provider’s staff cannot be recipients.
The first email covers the month in progress. It is sent from 06:00 on the 2nd of the next month in your organisation’s time zone, which gives the vendors’ bills a day to arrive.
Each month is sent once. Recipients are checked again at every send, so someone who has left the organisation stops receiving it. If a send fails, it is tried again on the next hourly run, up to three times a month, and the panel shows why.
The email carries the headline figures and the top findings, with the full report attached as PDF and HTML.
Use the API
Section titled “Use the API”Download the report for the environment your session or API key resolves to:
curl -H "Authorization: Bearer $SHIELD360_API_KEY" \ "https://app.shield360.ai/api/reports/ai-spend-risk?format=pdf&period=month&month=2026-10" \ -o ai-spend-risk-2026-10.pdf| Parameter | Values |
|---|---|
format | pdf (default) or html |
period | last30 (default) or month |
month | YYYY-MM, with period=month |
tz | An IANA time zone; defaults to the organisation’s |
GET and PUT /api/reports/ai-spend-risk/schedule read and change the monthly email ({ "enabled": true, "recipients": ["cfo@example.com"] }). Malformed input returns 400.