Data Processing & Subprocessors

For accounts on the hosted deployment, you are the controller of the data you submit and the deployment's operator is the processor: the software runs on the operator's infrastructure, under the operator's configuration, and the third parties below are that deployment's subprocessors. The software itself phones no home beyond what is listed here — there is no hidden analytics or telemetry endpoint in the code.

The third parties below are contacted by the deployment, on your behalf, when a scan runs. Each receives only what is listed against it.

Each of these receives only what is listed. None receives your account details, your users, or the contents of your findings.

A disclosure worth understanding: checking whether an undeclared import exists publicly means sending that package name to the npm or PyPI registry. For an internal package name, that discloses the name to the registry operator. The check can be turned off entirely with PHANTOM_DEP_REGISTRY_CHECK=off, in which case undeclared imports are still reported but without an existence verdict.

Alert delivery sends findings to a destination you choose. Nothing is sent until you configure one, and what is sent is the alert payload — asset, score, and finding titles and severities. Supported formats are generic webhook, Slack, PagerDuty, Datadog and Splunk. Whichever you point at, that vendor becomes a subprocessor of yours for the contents of those alerts.

The red-team suite is the other case: it sends adversarial probe prompts to a model endpoint you nominate and have proven you control. Those prompts are the inert canary payloads described in the AI Usage Policy, and the responses are stored truncated as evidence.

This application does not send your data to any AI model. It analyses AI systems and it can red-team a model endpoint you own, but it does not itself use a model to produce findings, write remediation, or process your data. Every finding is produced by deterministic rules in the source.

That is stated here rather than in marketing copy because it is the kind of claim people reasonably assume is untrue, and because it constrains what future changes may do without updating this page. See the AI Usage Policy.

The services above are operated in various jurisdictions, principally the United States. Because your deployment contacts them directly, any transfer is one your deployment makes — assess it against your own obligations. The data involved is limited to what the table lists: package names, CVE identifiers and domain names. No personal data is transmitted to any of them by this software.

This list is accurate for the code it ships with. It will become wrong the first time a scanner gains a new data source and nobody updates this page — so if you are operating a deployment for others, treat “add an outbound request” and “update this page” as the same change.