Legal

Privacy Policy

Effective {{EFFECTIVE}} · Last updated {{EFFECTIVE}}
Shield 360 is a product of ThinkFleet ("Company", "we", "us", "our").

Shield 360 measures what an organization spends on AI, what its AI tools and agents do, and what data they touch. It runs on machines belonging to people who did not choose to install it, so this policy is specific about what is collected, what is not, how long it is kept, and who decides.

1. Who this policy covers, and our role

Shield 360 creates three different relationships, and we treat each separately:

You are	Examples	Our role
A visitor or prospect	You read shield360.ai, took the assessment, or booked a call.	Controller
A customer user	You sign in to Shield 360 as an administrator or user of a customer organization.	Controller of your account data
A monitored individual	Your employer (our customer) installed the Shield 360 collector on a machine you use, or its applications send telemetry to Shield 360.	Processor (a "service provider" under US state law). Your employer is the controller and decides what is collected, for how long, and who may see it.

Self-hosted deployments send us nothing. Where a customer runs Shield 360 in its own environment, the data it collects stays in that customer's databases; ThinkFleet has no access to it and is not its processor. Sections 4 to 6 describe what the software does in any deployment; the rest of this policy applies to the hosted service and to this website.

If your employer deployed Shield 360

Your employer, not ThinkFleet, decides whether Shield 360 runs on your machine and what it records. Requests to see, correct or delete that data go to your employer. If you contact us, we will tell you which organization controls it and pass your request to them. We will not hand the data to you ourselves, because we cannot verify that the person asking is the person it is about.

2. Website visitors and prospects

Analytics. This site uses Google Analytics 4 to count visits and see which pages are read. See Section 11.
The AI governance assessment. Your answers are scored in your browser and are kept only in the link on your screen. We receive an event saying an assessment started or finished, with the overall score, band, role and the number of "Not sure" answers, but never your individual answers. We also receive an event when you finish each of its six areas, naming the area only. An unfinished assessment is saved in your browser's local storage, on your device only, so you can come back to it.
Emailing your assessment report. If you ask us to email your report, we receive your email address, the company name you enter and your result. We keep your address, company, role, overall score, band and three weakest areas, and whether you asked for a reminder or agreed to be contacted. We never keep your individual answers. We use this to send the report, the single 90-day reminder if you asked for one, and, only if you ticked the box, to contact you about your result. Every email has an unsubscribe link. We delete these records two years after your request. The email is sent through Amazon SES.
Watching the tour and opening the live demo. When you fill in the form on the demo page, we receive and keep your name, work email, company, role and company size, the page or campaign that brought you, and when you agreed to be contacted. We send you a link to the live demo that works for seven days, and record when you open it. We use this to show you the tour for your role and to contact you about Shield360. We keep these records until you ask us to delete them. The live demo shows a fictional company and never your data.
Downloading the NAIC and ISO 42001 toolkits. When you fill in the form on the toolkit page, we receive and keep your name, work email, company, role, the toolkit you chose, the page or campaign that brought you, and when you agreed to be contacted. We use this to give you the download links, which work for one hour, and to contact you about the toolkits and Shield 360. We keep these records until you ask us to delete them. Your browser also keeps a note that you asked, which opens the online workbook.
The online workbook. What you type in the workbook is kept only in your browser's local storage and is never sent to us or anyone else. Backup files you save stay on your device. Clearing this site's data in your browser deletes the workbook.
Booking a call. Calls are booked through Cal.com, which receives the name, email address and notes you enter. If you book from the assessment, the notes are pre-filled with your score, the areas where you scored lowest and your results link. You can see and edit them before you confirm.
Email. If you write to us, we keep the correspondence to answer you and to follow up about Shield 360.
Fonts. Pages load typefaces from Google Fonts, which receives your IP address and browser details when the page loads.

3. Customer users

Account information: name, email address and organization. On the hosted service, sign-in and organization provisioning run on Base316, ThinkFleet's own identity and billing platform. Where your organization federates sign-in, your own identity provider authenticates you over OIDC and we receive the identity claims it releases.
Billing information: subscriptions run on Base316, and payments are processed by Stripe. We do not receive or store full payment card numbers.
Configuration: projects, roles, collection and data protection policies, retention settings, legal holds, and the connections you set up.
Activity: the audit log records your changes, your reads of sensitive data and your sign-ins, with timestamp, outcome and source IP address.
Support communications: messages and attachments you send us.

4. What the software collects for a customer

The collector agent runs on enrolled machines. It reads information the operating system already keeps and files that AI tools write about themselves. Applications a customer instruments send telemetry over OpenTelemetry. The customer's configuration decides which of the following are collected. Default retention on the hosted service is enforced by the database itself: when a horizon passes, the rows are deleted.

Category	What it contains	Kept for
AI usage	Tool, provider, model, token counts, cost, the authentication type the tool reported, a key hint (a provider prefix and the last four characters, never the key), timestamps, the machine and the operating-system user account.	730 days
Coding-agent sessions	What Claude Code, Cursor or Codex did during a session: the steps, the tools it called, file paths and commands, and, when content capture is set to full, the text of prompts, responses and tool inputs and outputs. See the note below.	30 days
AI session output	Lines added and removed per session, the repository path and its remote where resolvable.	730 days
Application activity	Which application was in front (its identity and display name) and for how long. Never window titles or document names.	365 days; fine-grained spans 90 days
Browsing off by default	Registrable domain, browser, timestamp and visit count. The full address is stored only where an administrator has chosen the full level. Never page titles.	30 days
Network observations off by default	Which process reached an AI provider directly: process name, the provider host, the machine and user account, first and last seen. Connection tables only, never packet contents.	365 days
Machine state	Hostname, operating system, architecture, agent version, timezone, awake, asleep or locked; heartbeats showing the collector is reporting.	90 days; heartbeats 30 days
Data protection events	That a rule fired, which rule, the outcome (blocked, redacted or warned), the file or command involved, and a masked form and one-way fingerprint of the value detected, never the value itself.	7 years
Prompt-injection detections	Which rule matched, a score, where the content came from (a prompt, a tool result or description, a file, a served context), the agent or tool involved and the outcome. A short masked excerpt is kept only when content capture is set to full; the content itself is never stored.	7 years
AI code findings	Where a customer connects GitHub: the rule, file path, line and a short code excerpt of up to 200 characters.	365 days
Application telemetry	Traces, metrics, logs and exceptions that the customer's own applications send, which may include prompts and responses if the customer's instrumentation records them.	30 days for traces
Content the customer stores	Prompts in Prompt Hub, contexts and documents, secrets in Vault (encrypted), agent memory read from memory stores the customer connects, and agent grants and activity.	Until the customer deletes it, or the account ends (Section 9)
Alerts	Alerts raised for the customer (for example a spend threshold or a blocked prompt), who acknowledged them, and delivery status to the customer's channels. Channel credentials are encrypted and never shown again.	90 days after an alert closes
Agent test runs	The customer's test suites and each run's results and transcripts, including the tool calls an agent attempted. Transcripts are masked for secrets and personal data before they are stored.	Until the customer deletes them, or the account ends (Section 9)
Audit log	Changes, reads of sensitive data and sign-ins in the customer's organization.	For the life of the account; never edited

Prompt content, and who decides

Data protection works by reading a prompt at the point of use, before it is sent, so that a secret or personal data can be blocked or redacted. In the default full capture mode, the prompt and response text of coding-agent sessions is also stored, for 30 days, so that an administrator can see what was stopped and why. The customer's administrator can set capture to metadata_only or minimal, for the whole organization or for named people; a machine's own setting can make capture stricter, never wider. In the stricter modes, prompts are still checked on the machine but their text is not stored.

Records under a legal hold that the customer has placed are copied before their horizon passes and kept until the customer releases the hold. The hosted service's defaults are shown above; a self-hosted customer controls its own storage.

5. What the software never collects

These are properties of how Shield 360 is built, not settings:

No keystrokes, mouse activity, screenshots or screen recording.
No window titles and no document names in application activity: "Microsoft Word", not the name of the file.
No page titles. Browsing, when enabled, stores a domain unless the full level is chosen.
No packet capture, no proxy and no root certificate. Nothing is placed in the path of a request and no TLS connection is weakened in order to be read.
No full API keys observed on machines. Only a hint: the provider prefix and the last four characters.

Collection can only be narrowed for an individual

Browsing and network observation are off until an administrator enables them for the whole organization. A per-person setting can lower what is collected for that person; it cannot raise it above the organization's level, and data above a person's level is dropped when it arrives. Turning the lens onto one named individual is not something the system can express.

6. How we use information

To provide the service: attribute cost, discover AI use, apply the customer's data protection policy, run the evaluations and connections the customer configures, and produce reports and evidence exports.
To operate and secure it: monitor performance, fix faults, prevent abuse, and keep the audit log.
To communicate: service messages about your account, and replies to people who contact us or book a call.
To comply with law and enforce our terms.

We do not sell personal information and do not share it for cross-context behavioral advertising. We do not use customer data, including prompts, to train AI models, and data collected for one customer is never used to produce insights for another. We do not create aggregated "threat" or "telemetry" datasets from customer data for our own use.

7. Who we share it with

Sub-processors that host and operate the service, listed with their purpose and location on our sub-processors page.
Services the customer chooses to connect. When a customer connects a model provider for evaluations or Openground, a memory store, GitHub, an identity provider or an AI provider's billing API, Shield 360 exchanges data with that service on the customer's instruction and under the customer's own account. For example, running an evaluation sends the trace being scored (its prompt, response and context) to the model provider the customer selected, and an agent test sends its test prompts to the model or endpoint under test. Those services act under their agreements with the customer.
Destinations the customer chooses for alerts and security events. Alerts go to the Slack, Microsoft Teams, PagerDuty, webhook or email destinations the customer configures; alert emails are sent through Amazon Web Services. Security events exported to a customer's SIEM (such as Splunk, Microsoft Sentinel, Datadog, Elastic, the customer's own S3 bucket or syslog server) carry masked forms and fingerprints, never prompt text, detected values or secrets.
Legal requirements: where compelled by law. Where the law allows, we will tell the customer first and direct the request to them.
Business transfers: in a merger, acquisition or sale of assets, subject to this policy, with notice to customers.

8. Where it is stored, and how it is protected

The hosted service stores data in the United States, in Amazon Web Services' us-east-2 region. If you are outside the United States, your information is transferred there, and we rely on the European Commission's Standard Contractual Clauses, with the UK addendum, where a transfer mechanism is required.

Encryption in transit with TLS 1.2 or later.
Account databases, object storage and backups encrypted at rest with AES-256.
Provider credentials and secrets additionally encrypted at the application level with AES-256-GCM.
Role-based access control, enforced on the server for every request and scoped by organization, project and environment.
A separate database for each customer environment, rather than a shared table with a tenant column.
An audit log of changes, reads of sensitive data and sign-ins, exportable for evidence.

If a breach affects personal information we process for a customer, we will notify that customer without undue delay and in any case within 72 hours of confirming it, with what we know and what we are doing, and we will notify authorities and individuals where the law requires.

9. Retention, export and deletion

Service data is kept for the periods in Section 4, then deleted by the database.
A customer's organization owner can export the organization's data at any time. Secrets are exported by name only.
When an organization is closed or its subscription ends, its data is kept read-only for 30 days so it can be exported, then deleted. Data under an active legal hold is kept until the hold is released. We keep a minimal record that the deletion happened.
Account information is kept while the account is active and for as long afterwards as we need it for legal, tax and accounting obligations.
Website analytics are kept for Google Analytics' configured retention period, and correspondence for as long as the conversation is relevant.

10. Your rights

Depending on where you live, you may have the right to know what we hold about you, to access, correct, delete or port it, to object to or restrict its processing, to withdraw consent, and to appeal a decision we make about your request. We will not treat you differently for exercising them.

Visitors and customer users: email privacy@thinkfleet.ai. We verify requests against the email address we hold, accept requests from an authorized agent with your written permission, and respond within 30 days (45 days where California law applies). If we decline, we will say why and how to appeal.
Monitored individuals: see the note in Section 1. We help our customers answer these requests, and they can export or delete a person's data from the product.

10.1 California and other US states

In the past 12 months we have collected identifiers (name, email, IP address), internet activity (website analytics), professional information (organization and role), and, as a service provider for customers, the categories in Section 4. We collect them for the purposes in Section 6 and disclose them only as in Section 7. We do not sell or share personal information, and we do not knowingly collect personal information of anyone under 16. We honor Global Privacy Control signals as an opt-out on this website.

10.2 European Economic Area, United Kingdom and Switzerland

Where we are controller, we rely on performance of a contract (customer accounts), our legitimate interests in running, securing and marketing the service (analytics where consent is not required, correspondence, security logs), consent (analytics cookies where required), and legal obligations. You may complain to your data protection authority. Where we act as a processor for an employer, that employer is responsible for its own lawful basis for monitoring, including, where required, consulting a works council or equivalent body and completing an impact assessment before deployment. A data processing agreement is available to customers on request.

11. Cookies and analytics

The Shield 360 application uses only cookies and local storage that are strictly necessary for sign-in, session management and security, plus storage that remembers interface preferences. It carries no analytics or advertising tags.

This website uses Google Analytics 4, which sets Google's analytics cookies and sends Google your IP address, the pages you view, and device and browser information. We use it only to see which parts of the site are useful; we do not use it to identify you, we do not combine it with anything else we hold, and Google signals and advertising features are off.

In the EEA, the UK and Switzerland, analytics cookies are not set unless you accept them in the banner.
If your browser sends a Global Privacy Control signal, analytics cookies are not set, wherever you are.
You can change your choice at any time with the "Cookie settings" link in the site footer.

We run no advertising or remarketing tags.

12. Children

Shield 360 is a business service and is not directed to anyone under 18. We do not knowingly collect information from children, and if we learn that we have, we delete it.

13. Changes to this policy

We will update the date above when this policy changes. We will tell customers about material changes by email or in the application at least 30 days before they take effect.